andre/crewAI
1
0
Fork 0
mirror of https://github.com/crewAIInc/crewAI.git synced 2026-07-27 09:45:16 +00:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

docs(llms): refresh model availability guidance (#6676)
Some checks are pending
Vulnerability Scan / pip-audit (push) Waiting to run
Details
CodeQL Advanced / Analyze (actions) (push) Waiting to run
Details
CodeQL Advanced / Analyze (python) (push) Waiting to run
Details
Check Documentation Broken Links / Check broken links (push) Waiting to run
Details

Browse Source 
* docs(llms): refresh model availability guidance

* docs: clarify structured output support

* docs: address LLM guide review feedback

* docs: refresh streaming model examples
This commit is contained in:
Ossama Alami
2026-07-26 16:13:36 -07:00
committed by GitHub
parent 1870b444e7
commit daa7019898
4 changed files with 319 additions and 1004 deletions
Download Patch File Download Diff File Expand all files Collapse all files

339
docs/edge/ar/concepts/llms.mdx
View File

@@ -22,7 +22,7 @@ mode: "wide"
تحدد نافذة السياق مقدار النص الذي يمكن لـ LLM معالجته في وقت واحد. النوافذ الأكبر (مثل 128K رمز) تتيح سياقًا أكثر لكنها قد تكون أكثر تكلفة وأبطأ.
</Card>
<Card title="درجة الحرارة" icon="temperature-three-quarters">
تتحكم درجة الحرارة (0.0 إلى 1.0) في عشوائية الاستجابة. القيم المنخفضة (مثل 0.2) تنتج مخرجات أكثر تركيزًا وحتمية، بينما القيم الأعلى (مثل 0.8) تزيد الإبداع والتنوع.
درجة الحرارة هي أداة للتحكم في أخذ العينات تدعمها بعض النماذج. تجعل القيم المنخفضة أخذ العينات أكثر تركيزًا عمومًا، بينما تزيد القيم الأعلى التباين. تتجاهل بعض نماذج الاستدلال الأحدث هذا المعامل أو توقف دعمه أو ترفضه، لذا راجع وثائق النموذج المحدد قبل ضبطه.
</Card>
<Card title="اختيار المزود" icon="server">
يقدم كل مزود LLM (مثل OpenAI و Anthropic و Google) نماذج مختلفة بقدرات وأسعار وميزات متفاوتة. اختر بناءً على احتياجاتك من الدقة والسرعة والتكلفة.
@@ -38,7 +38,7 @@ mode: "wide"
أبسط طريقة للبدء. عيّن النموذج في بيئتك مباشرة، من خلال ملف `.env` أو في كود تطبيقك. إذا استخدمت `crewai create` لبدء مشروعك، سيكون مُعيّنًا بالفعل.
```bash .env
MODEL=model-id # e.g. gpt-4o, gemini-2.0-flash, claude-3-sonnet-...
MODEL=provider/model-id # e.g. openai/gpt-5.6-terra
# Be sure to set your API keys here too. See the Provider
# section below.
@@ -57,7 +57,7 @@ mode: "wide"
goal: Conduct comprehensive research and analysis
backstory: A dedicated research professional with years of experience
verbose: true
llm: provider/model-id # e.g. openai/gpt-4o, google/gemini-2.0-flash, anthropic/claude...
llm: provider/model-id # e.g. anthropic/claude-sonnet-4-6
# (see provider configuration examples below for more)
```
@@ -76,32 +76,24 @@ mode: "wide"
from crewai import LLM
# Basic configuration
llm = LLM(model="model-id-here") # gpt-4o, gemini-2.0-flash, anthropic/claude...
llm = LLM(model="provider/model-id") # e.g. gemini/gemini-3.6-flash
# Advanced configuration with detailed parameters
llm = LLM(
model="model-id-here", # gpt-4o, gemini-2.0-flash, anthropic/claude...
temperature=0.7, # Higher for more creative outputs
timeout=120, # Seconds to wait for response
max_tokens=4000, # Maximum length of response
top_p=0.9, # Nucleus sampling parameter
frequency_penalty=0.1 , # Reduce repetition
presence_penalty=0.1, # Encourage topic diversity
model="provider/model-id",
timeout=120,
max_tokens=4000,
response_format={"type": "json"}, # For structured outputs
seed=42 # For reproducible results
)
```
<Info>
شرح المعاملات:
- `temperature`: تتحكم في العشوائية (0.0-1.0)
- `timeout`: أقصى وقت انتظار للاستجابة
- `max_tokens`: تحدد طول الاستجابة
- `top_p`: بديل لدرجة الحرارة للعينات
- `frequency_penalty`: تقلل تكرار الكلمات
- `presence_penalty`: تشجع موضوعات جديدة
- `response_format`: تحدد هيكل المخرجات
- `seed`: تضمن مخرجات متسقة
عناصر التحكم في أخذ العينات مثل `temperature` و`top_p`، ومعاملات العقوبة، وأسماء حدود الرموز، وعناصر التحكم في الاستدلال خاصة بكل نموذج. أضفها فقط عندما يدعمها المزود والنموذج المحددان. راجع أمثلة المزودين أدناه ووثائق النموذج لدى المزود.
</Info>
</Tab>
</Tabs>
@@ -120,6 +112,10 @@ mode: "wide"
يدعم CrewAI العديد من مزودي LLM، كل منهم يقدم ميزات فريدة وطرق مصادقة وقدرات نماذج.
في هذا القسم، ستجد أمثلة مفصلة تساعدك في اختيار وإعداد وتحسين LLM الأنسب لاحتياجات مشروعك.
<Warning>
يتغير توفر النماذج باستمرار وقد يختلف حسب الحساب والمنطقة والمنصة السحابية. تستخدم الأمثلة أدناه نماذج متاحة وقت كتابة هذا الدليل، لكنها ليست قوائم دعم شاملة. قبل النشر، تحقق من معرّف النموذج وحالة دورة حياته في كتالوج المزود المرتبط.
</Warning>
<AccordionGroup>
<Accordion title="OpenAI">
يوفر CrewAI تكاملًا أصليًا مع OpenAI من خلال OpenAI Python SDK.
@@ -137,10 +133,10 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
api_key="your-api-key", # Or set OPENAI_API_KEY
temperature=0.7,
max_tokens=4000
reasoning_effort="medium",
max_completion_tokens=4000
)
```
@@ -149,25 +145,16 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
api_key="your-api-key",
base_url="https://api.openai.com/v1", # Optional custom endpoint
organization="org-...", # Optional organization ID
project="proj_...", # Optional project ID
temperature=0.7,
max_tokens=4000,
max_completion_tokens=4000, # For newer models
top_p=0.9,
frequency_penalty=0.1,
presence_penalty=0.1,
stop=["END"],
seed=42, # For reproducible outputs
max_completion_tokens=4000,
reasoning_effort="medium",
stream=True, # Enable streaming
timeout=60.0, # Request timeout in seconds
max_retries=3, # Maximum retry attempts
logprobs=True, # Return log probabilities
top_logprobs=5, # Number of most likely tokens
reasoning_effort="medium" # For o1 models: low, medium, high
max_retries=3 # Maximum retry attempts
)
```
@@ -182,7 +169,7 @@ mode: "wide"
summary: str
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
)
```
@@ -191,30 +178,15 @@ mode: "wide"
- `OPENAI_BASE_URL`: عنوان URL مخصص لـ OpenAI API (اختياري)
**الميزات:**
- دعم استدعاء الدوال الأصلي (باستثناء نماذج o1)
- دعم أصلي لاستدعاء الدوال
- مخرجات منظمة مع JSON schema
- دعم البث للاستجابات في الوقت الفعلي
- تتبع استخدام الرموز
- دعم تسلسلات التوقف (باستثناء نماذج o1)
- عناصر تحكم في التوليد خاصة بالمزود
- احتمالات السجل لرؤى على مستوى الرموز
- التحكم في جهد الاستدلال لنماذج o1
- التحكم في جهد الاستدلال للنماذج المتوافقة
**النماذج المدعومة:**
| النموذج | نافذة السياق | الأفضل لـ |
|---------------------|------------------|-----------------------------------------------|
| gpt-4.1 | 1M tokens | أحدث نموذج بقدرات محسّنة |
| gpt-4.1-mini | 1M tokens | إصدار فعال بسياق كبير |
| gpt-4.1-nano | 1M tokens | متغير فائق الكفاءة |
| gpt-4o | 128,000 tokens | محسّن للسرعة والذكاء |
| gpt-4o-mini | 200,000 tokens | فعال من حيث التكلفة بسياق كبير |
| gpt-4-turbo | 128,000 tokens | المحتوى الطويل، تحليل المستندات |
| gpt-4 | 8,192 tokens | مهام الدقة العالية، الاستدلال المعقد |
| o1 | 200,000 tokens | الاستدلال المتقدم، حل المشكلات المعقدة |
| o1-preview | 128,000 tokens | معاينة قدرات الاستدلال |
| o1-mini | 128,000 tokens | نموذج استدلال فعال |
| o3-mini | 200,000 tokens | نموذج استدلال خفيف |
| o4-mini | 200,000 tokens | استدلال فعال من الجيل التالي |
تضيف OpenAI نماذج جديدة وتسحب snapshots قديمة بانتظام. راجع [كتالوج نماذج OpenAI](https://developers.openai.com/api/docs/models) للحصول على معرّفات النماذج الحالية ونوافذ السياق وتوافق endpoints ومعلومات دورة الحياة.
**Responses API:**
@@ -276,14 +248,7 @@ mode: "wide"
)
```
جميع النماذج المدرجة هنا https://llama.developer.meta.com/docs/models/ مدعومة.
| معرّف النموذج | طول سياق الإدخال | طول سياق المخرجات | وسائط الإدخال | وسائط المخرجات |
| --- | --- | --- | --- | --- |
| `meta_llama/Llama-4-Scout-17B-16E-Instruct-FP8` | 128k | 4028 | نص، صورة | نص |
| `meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8` | 128k | 4028 | نص، صورة | نص |
| `meta_llama/Llama-3.3-70B-Instruct` | 128k | 4028 | نص | نص |
| `meta_llama/Llama-3.3-8B-Instruct` | 128k | 4028 | نص | نص |
راجع [نظرة عامة على نماذج Meta Llama](https://ai.meta.com/llama/get-started/) للتعرّف على عائلات النماذج والوسائط وإرشادات حدود السياق الحالية.
**ملاحظة:** يستخدم هذا المزود LiteLLM. أضفه كتبعية لمشروعك:
```bash
@@ -353,7 +318,7 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="anthropic/claude-3-5-sonnet-20241022",
model="anthropic/claude-sonnet-4-6",
api_key="your-api-key", # Or set ANTHROPIC_API_KEY
max_tokens=4096 # Required for Anthropic
)
@@ -364,12 +329,10 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="anthropic/claude-3-5-sonnet-20241022",
model="anthropic/claude-sonnet-4-6",
api_key="your-api-key",
base_url="https://api.anthropic.com", # Optional custom endpoint
temperature=0.7,
max_tokens=4096, # Required parameter
top_p=0.9,
stop_sequences=["END", "STOP"], # Anthropic uses stop_sequences
stream=True, # Enable streaming
timeout=60.0, # Request timeout in seconds
@@ -377,7 +340,7 @@ mode: "wide"
)
```
**التفكير الموسّع (Claude Sonnet 4 وما بعده):**
**التفكير الموسّع:**
يدعم CrewAI ميزة التفكير الموسّع من Anthropic، التي تتيح لـ Claude التفكير في المشكلات بطريقة أكثر شبهًا بالبشر قبل الاستجابة. مفيد بشكل خاص لمهام الاستدلال والتحليل وحل المشكلات المعقدة.
@@ -386,14 +349,14 @@ mode: "wide"
# Enable extended thinking with default settings
llm = LLM(
model="anthropic/claude-sonnet-4",
model="anthropic/claude-sonnet-4-6",
thinking={"type": "enabled"},
max_tokens=10000
)
# Configure thinking with budget control
llm = LLM(
model="anthropic/claude-sonnet-4",
model="anthropic/claude-sonnet-4-6",
thinking={
"type": "enabled",
"budget_tokens": 5000 # Limit thinking tokens
@@ -406,9 +369,7 @@ mode: "wide"
- `type`: عيّن إلى `"enabled"` لتفعيل وضع التفكير الموسّع
- `budget_tokens` (اختياري): أقصى رموز للتفكير (يساعد في التحكم بالتكاليف)
**النماذج التي تدعم التفكير الموسّع:**
- `claude-sonnet-4` والنماذج الأحدث
- `claude-3-7-sonnet` (مع قدرات التفكير الموسّع)
تختلف أوضاع التفكير والمعاملات المقبولة بين أجيال Claude. تحقق من قدرات النموذج المحدد قبل تفعيل `thinking`.
**متى تستخدم التفكير الموسّع:**
- الاستدلال المعقد وحل المشكلات متعددة الخطوات
@@ -424,7 +385,7 @@ mode: "wide"
**الميزات:**
- دعم استخدام الأدوات الأصلي لنماذج Claude 3+
- دعم التفكير الموسّع لـ Claude Sonnet 4+
- دعم التفكير الموسّع لنماذج Claude المتوافقة
- دعم البث للاستجابات في الوقت الفعلي
- معالجة تلقائية لرسائل النظام
- تسلسلات التوقف للتحكم في المخرجات
@@ -438,20 +399,7 @@ mode: "wide"
- يجب أن تكون الرسالة الأولى من المستخدم (يتم التعامل معها تلقائيًا)
- يجب أن تتناوب الرسائل بين المستخدم والمساعد
**النماذج المدعومة:**
| النموذج | نافذة السياق | الأفضل لـ |
|------------------------------|------------------|-----------------------------------------------|
| claude-sonnet-4 | 200,000 tokens | الأحدث مع قدرات التفكير الموسّع |
| claude-3-7-sonnet | 200,000 tokens | الاستدلال المتقدم والمهام الوكيلية |
| claude-3-5-sonnet-20241022 | 200,000 tokens | أحدث Sonnet بأفضل أداء |
| claude-3-5-haiku | 200,000 tokens | نموذج سريع وصغير للاستجابات السريعة |
| claude-3-opus | 200,000 tokens | الأكثر قدرة للمهام المعقدة |
| claude-3-sonnet | 200,000 tokens | توازن بين الذكاء والسرعة |
| claude-3-haiku | 200,000 tokens | الأسرع للمهام البسيطة |
| claude-2.1 | 200,000 tokens | سياق موسّع، هلوسات أقل |
| claude-2 | 100,000 tokens | نموذج متعدد الاستخدامات |
| claude-instant | 100,000 tokens | سريع وفعال من حيث التكلفة للمهام اليومية |
راجع [نظرة عامة على نماذج Anthropic](https://platform.claude.com/docs/en/about-claude/models/overview) للحصول على معرّفات النماذج وقدراتها الحالية، وراجع [جدول إيقاف النماذج](https://platform.claude.com/docs/en/about-claude/model-deprecations) قبل تثبيت نموذج في الإنتاج.
**ملاحظة:** لاستخدام Anthropic، ثبّت التبعيات المطلوبة:
```bash
@@ -483,9 +431,8 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.0-flash",
model="gemini/gemini-3.6-flash",
api_key="your-api-key", # Or set GOOGLE_API_KEY/GEMINI_API_KEY
temperature=0.7
)
```
@@ -494,11 +441,8 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.5-flash",
model="gemini/gemini-3.6-flash",
api_key="your-api-key",
temperature=0.7,
top_p=0.9,
top_k=40, # Top-k sampling parameter
max_output_tokens=8192,
stop_sequences=["END", "STOP"],
stream=True, # Enable streaming
@@ -524,8 +468,7 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.0-flash",
temperature=0.7
model="gemini/gemini-3.6-flash"
)
```
@@ -542,7 +485,7 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="gemini/gemini-1.5-pro",
model="gemini/gemini-3.6-flash",
project="your-gcp-project-id",
location="us-central1" # GCP region
)
@@ -555,7 +498,7 @@ mode: "wide"
- `GOOGLE_CLOUD_LOCATION`: موقع GCP (الافتراضي `us-central1`)
**الميزات:**
- دعم استدعاء الدوال الأصلي لنماذج Gemini 1.5+ و 2.x
- دعم أصلي لاستدعاء الدوال لنماذج Gemini المتوافقة
- دعم البث للاستجابات في الوقت الفعلي
- قدرات متعددة الوسائط (نص، صور، فيديو)
- إعداد إعدادات الأمان
@@ -563,41 +506,21 @@ mode: "wide"
- معالجة تلقائية لتعليمات النظام
- تتبع استخدام الرموز
**نماذج Gemini:**
| النموذج | نافذة السياق | الأفضل لـ |
|--------------------------------|-----------------|-------------------------------------------------------------------|
| gemini-2.5-flash | 1M tokens | التفكير التكيفي، كفاءة التكلفة |
| gemini-2.5-pro | 1M tokens | التفكير والاستدلال المحسّن، الفهم متعدد الوسائط |
| gemini-2.0-flash | 1M tokens | ميزات الجيل التالي، السرعة، التفكير |
| gemini-2.0-flash-thinking | 32,768 tokens | الاستدلال المتقدم مع عملية التفكير |
| gemini-2.0-flash-lite | 1M tokens | كفاءة التكلفة ووقت الاستجابة المنخفض |
| gemini-1.5-pro | 2M tokens | الأفضل أداءً، الاستدلال المنطقي، البرمجة |
| gemini-1.5-flash | 1M tokens | نموذج متعدد الوسائط متوازن، جيد لمعظم المهام |
| gemini-1.5-flash-8b | 1M tokens | الأسرع والأكثر كفاءة من حيث التكلفة |
| gemini-1.0-pro | 32,768 tokens | نموذج الجيل السابق |
تنشر Google معرّفات Gemini الحالية وقدراتها ومراحل دورة حياتها في [كتالوج نماذج Gemini](https://ai.google.dev/gemini-api/docs/models). تحقق من [جدول الإيقاف](https://ai.google.dev/gemini-api/docs/deprecations) قبل اختيار نموذج مستقر أو preview. وتستضيف Gemini API أيضًا [نماذج Gemma](https://ai.google.dev/gemma/docs).
**ملاحظة:** لاستخدام Google Gemini، ثبّت التبعيات المطلوبة:
```bash
uv add "crewai[google-genai]"
```
القائمة الكاملة للنماذج متاحة في [وثائق نماذج Gemini](https://ai.google.dev/gemini-api/docs/models).
</Accordion>
<Accordion title="Google (Vertex AI)">
احصل على بيانات الاعتماد من Google Cloud Console واحفظها في ملف JSON، ثم حمّلها بالكود التالي:
```python Code
import json
file_path = 'path/to/vertex_ai_service_account.json'
# Load the JSON file
with open(file_path, 'r') as file:
vertex_credentials = json.load(file)
# Convert the credentials to a JSON string
vertex_credentials_json = json.dumps(vertex_credentials)
صادِق باستخدام [بيانات الاعتماد التلقائية للتطبيق](https://cloud.google.com/docs/authentication/provide-credentials-adc)، ثم اضبط مزود Gemini الأصلي لاستخدام Vertex AI:
```toml .env
GOOGLE_GENAI_USE_VERTEXAI=true
GOOGLE_CLOUD_PROJECT=<your-project-id>
GOOGLE_CLOUD_LOCATION=<location>
```
مثال الاستخدام في مشروع CrewAI:
@@ -605,15 +528,15 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="gemini-1.5-pro-latest", # or vertex_ai/gemini-1.5-pro-latest
temperature=0.7,
vertex_credentials=vertex_credentials_json
model="gemini/gemini-3.6-flash"
)
```
**ملاحظة:** يستخدم هذا المزود LiteLLM. أضفه كتبعية لمشروعك:
تختلف إتاحة Vertex AI باختلاف المنطقة. استخدم [كتالوج نماذج Vertex AI](https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models) للتحقق من معرّف النموذج والمنطقة قبل النشر.
**ملاحظة:** يستخدم هذا المسار تكامل Gemini الأصلي في CrewAI. أضفه كتبعية لمشروعك:
```bash
uv add 'crewai[litellm]'
uv add "crewai[google-genai]"
```
</Accordion>
@@ -664,7 +587,7 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
model="bedrock/us.anthropic.claude-sonnet-4-6",
region_name="us-east-1"
)
```
@@ -674,7 +597,7 @@ mode: "wide"
from crewai import LLM
llm = LLM(
model="bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
model="bedrock/us.anthropic.claude-sonnet-4-6",
aws_access_key_id="your-access-key", # Or set AWS_ACCESS_KEY_ID
aws_secret_access_key="your-secret-key", # Or set AWS_SECRET_ACCESS_KEY
aws_session_token="your-session-token", # For temporary credentials
@@ -719,37 +642,9 @@ mode: "wide"
- يجب أن تكون الرسالة الأولى من المستخدم (يتم التعامل معها تلقائيًا)
- بعض النماذج (مثل Cohere) تتطلب أن تنتهي المحادثة برسالة المستخدم
[Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html) هو خدمة مُدارة توفر الوصول إلى نماذج أساسية متعددة من أبرز شركات الذكاء الاصطناعي عبر واجهة API موحدة.
| النموذج | نافذة السياق | الأفضل لـ |
|-------------------------|----------------------|-------------------------------------------------------------------|
| Amazon Nova Pro | حتى 300k tokens | أداء عالٍ، نموذج يوازن بين الدقة والسرعة والفعالية من حيث التكلفة عبر مهام متنوعة. |
| Amazon Nova Micro | حتى 128k tokens | نموذج نصي فقط عالي الأداء وفعال من حيث التكلفة ومحسّن لأقل وقت استجابة. |
| Amazon Nova Lite | حتى 300k tokens | معالجة متعددة الوسائط بأسعار معقولة للصور والفيديو والنص مع قدرات في الوقت الفعلي. |
| Claude 3.7 Sonnet | حتى 128k tokens | الأفضل أداءً للاستدلال المعقد والبرمجة ووكلاء الذكاء الاصطناعي |
| Claude 3.5 Sonnet v2 | حتى 200k tokens | نموذج متطور متخصص في هندسة البرمجيات والقدرات الوكيلية والتفاعل مع الحاسوب بتكلفة محسّنة. |
| Claude 3.5 Sonnet | حتى 200k tokens | نموذج عالي الأداء يقدم ذكاءً واستدلالًا فائقين عبر مهام متنوعة مع توازن مثالي بين السرعة والتكلفة. |
| Claude 3.5 Haiku | حتى 200k tokens | نموذج متعدد الوسائط سريع وصغير محسّن للاستجابات السريعة والتفاعلات الشبيهة بالبشر |
| Claude 3 Sonnet | حتى 200k tokens | نموذج متعدد الوسائط يوازن بين الذكاء والسرعة للنشر بكميات كبيرة. |
| Claude 3 Haiku | حتى 200k tokens | نموذج متعدد الوسائط صغير وسريع محسّن للاستجابات السريعة والتفاعلات المحادثية الطبيعية |
| Claude 3 Opus | حتى 200k tokens | أكثر النماذج متعددة الوسائط تقدمًا يتفوق في المهام المعقدة بالاستدلال الشبيه بالبشر والفهم السياقي الفائق. |
| Claude 2.1 | حتى 200k tokens | إصدار محسّن بنافذة سياق موسّعة وموثوقية محسّنة وهلوسات أقل لتطبيقات النصوص الطويلة وRAG |
| Claude | حتى 100k tokens | نموذج متعدد الاستخدامات يتفوق في الحوار المتقدم والمحتوى الإبداعي واتباع التعليمات الدقيقة. |
| Claude Instant | حتى 100k tokens | نموذج سريع وفعال من حيث التكلفة للمهام اليومية مثل الحوار والتحليل والتلخيص والأسئلة والأجوبة |
| Llama 3.1 405B Instruct | حتى 128k tokens | نموذج LLM متقدم لتوليد البيانات الاصطناعية والتقطير والاستدلال لروبوتات المحادثة والبرمجة والمهام المتخصصة. |
| Llama 3.1 70B Instruct | حتى 128k tokens | يدعم المحادثات المعقدة مع فهم سياقي فائق واستدلال وتوليد نص. |
| Llama 3.1 8B Instruct | حتى 128k tokens | نموذج متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| Llama 3 70B Instruct | حتى 8k tokens | يدعم المحادثات المعقدة مع فهم سياقي فائق واستدلال وتوليد نص. |
| Llama 3 8B Instruct | حتى 8k tokens | نموذج LLM متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| Titan Text G1 - Lite | حتى 4k tokens | نموذج خفيف وفعال من حيث التكلفة محسّن لمهام اللغة الإنجليزية والضبط الدقيق مع التركيز على التلخيص وتوليد المحتوى. |
| Titan Text G1 - Express | حتى 8k tokens | نموذج متعدد الاستخدامات لمهام اللغة العامة والمحادثة وتطبيقات RAG مع دعم الإنجليزية وأكثر من 100 لغة. |
| Cohere Command | حتى 4k tokens | نموذج متخصص في اتباع أوامر المستخدم وتقديم حلول عملية للمؤسسات. |
| Jurassic-2 Mid | حتى 8,191 tokens | نموذج فعال من حيث التكلفة يوازن بين الجودة والسعر لمهام اللغة المتنوعة مثل الأسئلة والأجوبة والتلخيص وتوليد المحتوى. |
| Jurassic-2 Ultra | حتى 8,191 tokens | نموذج لتوليد النص المتقدم والفهم، يتفوق في المهام المعقدة مثل التحليل وإنشاء المحتوى. |
| Jamba-Instruct | حتى 256k tokens | نموذج بنافذة سياق موسّعة محسّن لتوليد النص الفعال من حيث التكلفة والتلخيص والأسئلة والأجوبة. |
| Mistral 7B Instruct | حتى 32k tokens | نموذج LLM يتبع التعليمات ويكمل الطلبات ويولد نصًا إبداعيًا. |
| Mistral 8x7B Instruct | حتى 32k tokens | نموذج LLM بمعمارية MOE يتبع التعليمات ويكمل الطلبات ويولد نصًا إبداعيًا. |
| DeepSeek R1 | 32,768 tokens | نموذج استدلال متقدم |
تختلف إتاحة نماذج Amazon Bedrock ومعرّفاتها باختلاف المنطقة. استخدم مرجع
[النماذج والمناطق المدعومة](https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html)
لاختيار نموذج والتحقق من دعم Converse API.
**ملاحظة:** لاستخدام AWS Bedrock، ثبّت التبعيات المطلوبة:
```bash
@@ -806,81 +701,13 @@ mode: "wide"
مثال الاستخدام في مشروع CrewAI:
```python Code
llm = LLM(
model="nvidia_nim/meta/llama3-70b-instruct",
model="nvidia_nim/nvidia/nvidia-nemotron-3-ultra-550b-a55b",
temperature=0.7
)
```
يوفر Nvidia NIM مجموعة شاملة من النماذج لحالات الاستخدام المتنوعة، من المهام ذات الأغراض العامة إلى التطبيقات المتخصصة.
يتغير كتالوج NVIDIA NIM المستضاف باستمرار. استخدم [كتالوج نماذج NVIDIA NIM](https://build.nvidia.com/models) لاختيار endpoint حالي والتحقق من معرّف النموذج والوسائط وحدود السياق.
| النموذج | نافذة السياق | الأفضل لـ |
|-------------------------------------------------------------------------|----------------|-------------------------------------------------------------------|
| nvidia/mistral-nemo-minitron-8b-8k-instruct | 8,192 tokens | نموذج لغة صغير متطور يقدم دقة فائقة لروبوتات المحادثة والمساعدين الافتراضيين وتوليد المحتوى. |
| nvidia/nemotron-4-mini-hindi-4b-instruct | 4,096 tokens | نموذج لغة صغير ثنائي اللغة هندي-إنجليزي للاستدلال على الجهاز، مصمم خصيصًا للغة الهندية. |
| nvidia/llama-3.1-nemotron-70b-instruct | 128k tokens | مخصص لتعزيز فائدة الاستجابات |
| nvidia/llama3-chatqa-1.5-8b | 128k tokens | نموذج LLM متقدم لتوليد استجابات عالية الجودة ومدركة للسياق لروبوتات المحادثة ومحركات البحث. |
| nvidia/llama3-chatqa-1.5-70b | 128k tokens | نموذج LLM متقدم لتوليد استجابات عالية الجودة ومدركة للسياق لروبوتات المحادثة ومحركات البحث. |
| nvidia/vila | 128k tokens | نموذج رؤية-لغة متعدد الوسائط يفهم النص والصور والفيديو وينشئ استجابات غنية بالمعلومات |
| nvidia/neva-22 | 4,096 tokens | نموذج رؤية-لغة متعدد الوسائط يفهم النص والصور ويولد استجابات غنية بالمعلومات |
| nvidia/nemotron-mini-4b-instruct | 8,192 tokens | مهام ذات أغراض عامة |
| nvidia/usdcode-llama3-70b-instruct | 128k tokens | نموذج LLM متطور يجيب على استعلامات معرفة OpenUSD ويولد كود USD-Python. |
| nvidia/nemotron-4-340b-instruct | 4,096 tokens | ينشئ بيانات اصطناعية متنوعة تحاكي خصائص بيانات العالم الحقيقي. |
| meta/codellama-70b | 100k tokens | نموذج LLM قادر على توليد الكود من اللغة الطبيعية والعكس. |
| meta/llama2-70b | 4,096 tokens | نموذج لغة كبير متطور قادر على توليد النص والكود استجابة للمطالبات. |
| meta/llama3-8b-instruct | 8,192 tokens | نموذج LLM متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| meta/llama3-70b-instruct | 8,192 tokens | يدعم المحادثات المعقدة مع فهم سياقي فائق واستدلال وتوليد نص. |
| meta/llama-3.1-8b-instruct | 128k tokens | نموذج متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| meta/llama-3.1-70b-instruct | 128k tokens | يدعم المحادثات المعقدة مع فهم سياقي فائق واستدلال وتوليد نص. |
| meta/llama-3.1-405b-instruct | 128k tokens | نموذج LLM متقدم لتوليد البيانات الاصطناعية والتقطير والاستدلال لروبوتات المحادثة والبرمجة والمهام المتخصصة. |
| meta/llama-3.2-1b-instruct | 128k tokens | نموذج لغة صغير متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| meta/llama-3.2-3b-instruct | 128k tokens | نموذج لغة صغير متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| meta/llama-3.2-11b-vision-instruct | 128k tokens | نموذج لغة صغير متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| meta/llama-3.2-90b-vision-instruct | 128k tokens | نموذج لغة صغير متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| google/gemma-7b | 8,192 tokens | نموذج متطور لتوليد النص وفهمه وتحويله وتوليد الكود. |
| google/gemma-2b | 8,192 tokens | نموذج متطور لتوليد النص وفهمه وتحويله وتوليد الكود. |
| google/codegemma-7b | 8,192 tokens | نموذج متطور مبني على Gemma-7B من Google متخصص في توليد الكود وإكماله. |
| google/codegemma-1.1-7b | 8,192 tokens | نموذج برمجة متقدم لتوليد الكود وإكماله والاستدلال واتباع التعليمات. |
| google/recurrentgemma-2b | 8,192 tokens | نموذج لغة بمعمارية تكرارية جديدة لاستدلال أسرع عند توليد تسلسلات طويلة. |
| google/gemma-2-9b-it | 8,192 tokens | نموذج متطور لتوليد النص وفهمه وتحويله وتوليد الكود. |
| google/gemma-2-27b-it | 8,192 tokens | نموذج متطور لتوليد النص وفهمه وتحويله وتوليد الكود. |
| google/gemma-2-2b-it | 8,192 tokens | نموذج متطور لتوليد النص وفهمه وتحويله وتوليد الكود. |
| google/deplot | 512 tokens | نموذج فهم لغة بصرية بلقطة واحدة يترجم صور الرسوم البيانية إلى جداول. |
| google/paligemma | 8,192 tokens | نموذج لغة بصري بارع في استيعاب مدخلات النص والصور لإنتاج استجابات غنية بالمعلومات. |
| mistralai/mistral-7b-instruct-v0.2 | 32k tokens | نموذج LLM يتبع التعليمات ويكمل الطلبات ويولد نصًا إبداعيًا. |
| mistralai/mixtral-8x7b-instruct-v0.1 | 8,192 tokens | نموذج LLM بمعمارية MOE يتبع التعليمات ويكمل الطلبات ويولد نصًا إبداعيًا. |
| mistralai/mistral-large | 4,096 tokens | ينشئ بيانات اصطناعية متنوعة تحاكي خصائص بيانات العالم الحقيقي. |
| mistralai/mixtral-8x22b-instruct-v0.1 | 8,192 tokens | ينشئ بيانات اصطناعية متنوعة تحاكي خصائص بيانات العالم الحقيقي. |
| mistralai/mistral-7b-instruct-v0.3 | 32k tokens | نموذج LLM يتبع التعليمات ويكمل الطلبات ويولد نصًا إبداعيًا. |
| nv-mistralai/mistral-nemo-12b-instruct | 128k tokens | أكثر نموذج لغة تقدمًا للاستدلال والبرمجة والمهام متعددة اللغات؛ يعمل على وحدة GPU واحدة. |
| mistralai/mamba-codestral-7b-v0.1 | 256k tokens | نموذج للكتابة والتفاعل مع الكود عبر مجموعة واسعة من لغات البرمجة والمهام. |
| microsoft/phi-3-mini-128k-instruct | 128K tokens | نموذج LLM مفتوح خفيف ومتطور مع مهارات قوية في الرياضيات والاستدلال المنطقي. |
| microsoft/phi-3-mini-4k-instruct | 4,096 tokens | نموذج LLM مفتوح خفيف ومتطور مع مهارات قوية في الرياضيات والاستدلال المنطقي. |
| microsoft/phi-3-small-8k-instruct | 8,192 tokens | نموذج LLM مفتوح خفيف ومتطور مع مهارات قوية في الرياضيات والاستدلال المنطقي. |
| microsoft/phi-3-small-128k-instruct | 128K tokens | نموذج LLM مفتوح خفيف ومتطور مع مهارات قوية في الرياضيات والاستدلال المنطقي. |
| microsoft/phi-3-medium-4k-instruct | 4,096 tokens | نموذج LLM مفتوح خفيف ومتطور مع مهارات قوية في الرياضيات والاستدلال المنطقي. |
| microsoft/phi-3-medium-128k-instruct | 128K tokens | نموذج LLM مفتوح خفيف ومتطور مع مهارات قوية في الرياضيات والاستدلال المنطقي. |
| microsoft/phi-3.5-mini-instruct | 128K tokens | نموذج LLM خفيف متعدد اللغات يدعم تطبيقات الذكاء الاصطناعي في البيئات المحدودة بالكمون والذاكرة والحوسبة |
| microsoft/phi-3.5-moe-instruct | 128K tokens | نموذج LLM متقدم يعتمد على معمارية خليط الخبراء لتوليد محتوى فعال حوسبيًا |
| microsoft/kosmos-2 | 1,024 tokens | نموذج متعدد الوسائط رائد مصمم لفهم العناصر المرئية في الصور والاستدلال عليها. |
| microsoft/phi-3-vision-128k-instruct | 128k tokens | نموذج متعدد الوسائط مفتوح متطور يتفوق في الاستدلال عالي الجودة من الصور. |
| microsoft/phi-3.5-vision-instruct | 128k tokens | نموذج متعدد الوسائط مفتوح متطور يتفوق في الاستدلال عالي الجودة من الصور. |
| databricks/dbrx-instruct | 12k tokens | نموذج LLM للأغراض العامة بأداء متطور في فهم اللغة والبرمجة وRAG. |
| snowflake/arctic | 1,024 tokens | يقدم استدلالًا عالي الكفاءة لتطبيقات المؤسسات مع التركيز على توليد SQL والبرمجة. |
| aisingapore/sea-lion-7b-instruct | 4,096 tokens | نموذج LLM لتمثيل وخدمة التنوع اللغوي والثقافي لجنوب شرق آسيا |
| ibm/granite-8b-code-instruct | 4,096 tokens | نموذج LLM لبرمجة البرمجيات لتوليد الكود وإكماله وشرحه والتحويل متعدد الأدوار. |
| ibm/granite-34b-code-instruct | 8,192 tokens | نموذج LLM لبرمجة البرمجيات لتوليد الكود وإكماله وشرحه والتحويل متعدد الأدوار. |
| ibm/granite-3.0-8b-instruct | 4,096 tokens | نموذج لغة صغير متقدم يدعم RAG والتلخيص والتصنيف والكود والذكاء الاصطناعي الوكيلي |
| ibm/granite-3.0-3b-a800m-instruct | 4,096 tokens | نموذج خليط خبراء عالي الكفاءة لـ RAG والتلخيص واستخراج الكيانات والتصنيف |
| mediatek/breeze-7b-instruct | 4,096 tokens | ينشئ بيانات اصطناعية متنوعة تحاكي خصائص بيانات العالم الحقيقي. |
| upstage/solar-10.7b-instruct | 4,096 tokens | يتفوق في مهام NLP، خاصة في اتباع التعليمات والاستدلال والرياضيات. |
| writer/palmyra-med-70b-32k | 32k tokens | نموذج LLM رائد للاستجابات الدقيقة والمناسبة للسياق في المجال الطبي. |
| writer/palmyra-med-70b | 32k tokens | نموذج LLM رائد للاستجابات الدقيقة والمناسبة للسياق في المجال الطبي. |
| writer/palmyra-fin-70b-32k | 32k tokens | نموذج LLM متخصص في التحليل المالي وإعداد التقارير ومعالجة البيانات |
| 01-ai/yi-large | 32k tokens | نموذج قوي مدرب على الإنجليزية والصينية لمهام متنوعة بما في ذلك روبوتات المحادثة والكتابة الإبداعية. |
| deepseek-ai/deepseek-coder-6.7b-instruct | 2k tokens | نموذج برمجة قوي يقدم قدرات متقدمة في توليد الكود وإكماله وملء الفراغات |
| rakuten/rakutenai-7b-instruct | 1,024 tokens | نموذج LLM متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| rakuten/rakutenai-7b-chat | 1,024 tokens | نموذج LLM متطور مع فهم اللغة واستدلال فائق وتوليد النص. |
| baichuan-inc/baichuan2-13b-chat | 4,096 tokens | يدعم المحادثة بالصينية والإنجليزية والبرمجة والرياضيات واتباع التعليمات وحل الألغاز |
**ملاحظة:** يستخدم هذا المزود LiteLLM. أضفه كتبعية لمشروعك:
```bash
@@ -943,15 +770,12 @@ mode: "wide"
مثال الاستخدام في مشروع CrewAI:
```python Code
llm = LLM(
model="groq/llama-3.2-90b-text-preview",
model="groq/qwen/qwen3.6-27b",
temperature=0.7
)
```
| النموذج | نافذة السياق | الأفضل لـ |
|-------------------|------------------|--------------------------------------------|
| Llama 3.1 70B/8B | 131,072 tokens | مهام عالية الأداء بسياق كبير |
| Llama 3.2 Series | 8,192 tokens | مهام ذات أغراض عامة |
| Mixtral 8x7B | 32,768 tokens | أداء متوازن وسياق جيد |
تميز Groq بين نماذج production وpreview وتسحب معرّفات النماذج بانتظام. تحقق من [كتالوج نماذج Groq](https://console.groq.com/docs/models) و[صفحة الإيقاف](https://console.groq.com/docs/deprecations) قبل اختيار نموذج للإنتاج.
**ملاحظة:** يستخدم هذا المزود LiteLLM. أضفه كتبعية لمشروعك:
```bash
@@ -1033,11 +857,12 @@ mode: "wide"
مثال الاستخدام في مشروع CrewAI:
```python Code
llm = LLM(
model="llama-3.1-sonar-large-128k-online",
base_url="https://api.perplexity.ai/"
model="perplexity/sonar-pro"
)
```
راجع [كتالوج نماذج Perplexity](https://docs.perplexity.ai/getting-started/models) و[changelog](https://docs.perplexity.ai/docs/resources/changelog) للحصول على معرّفات النماذج الحالية وإشعارات الإيقاف.
**ملاحظة:** يستخدم هذا المزود LiteLLM. أضفه كتبعية لمشروعك:
```bash
uv add 'crewai[litellm]'
@@ -1073,17 +898,12 @@ mode: "wide"
مثال الاستخدام في مشروع CrewAI:
```python Code
llm = LLM(
model="sambanova/Meta-Llama-3.1-8B-Instruct",
model="sambanova/Meta-Llama-3.3-70B-Instruct",
temperature=0.7
)
```
| النموذج | نافذة السياق | الأفضل لـ |
|--------------------|------------------------|----------------------------------------------|
| Llama 3.1 70B/8B | حتى 131,072 tokens | مهام عالية الأداء بسياق كبير |
| Llama 3.1 405B | 8,192 tokens | أداء عالٍ وجودة مخرجات |
| Llama 3.2 Series | 8,192 tokens | مهام عامة ومتعددة الوسائط |
| Llama 3.3 70B | حتى 131,072 tokens | أداء عالٍ وجودة مخرجات |
| Qwen2 familly | 8,192 tokens | أداء عالٍ وجودة مخرجات |
قد تتغير النماذج المستضافة في SambaNova Cloud بصورة مستقلة عن CrewAI. استعلم من [models endpoint](https://docs.sambanova.ai/docs/api-reference/models/get-environments-available-model-list-metadata) وراجع [دليل الإيقاف](https://docs.sambanova.ai/docs/en/models/deprecations) قبل النشر.
**ملاحظة:** يستخدم هذا المزود LiteLLM. أضفه كتبعية لمشروعك:
```bash
@@ -1101,7 +921,7 @@ mode: "wide"
مثال الاستخدام في مشروع CrewAI:
```python Code
llm = LLM(
model="cerebras/llama3.1-70b",
model="cerebras/gpt-oss-120b",
temperature=0.7,
max_tokens=8192
)
@@ -1115,6 +935,8 @@ mode: "wide"
- دعم نوافذ سياق طويلة
</Info>
راجع [كتالوج نماذج Cerebras](https://inference-docs.cerebras.ai/models/overview) و[إشعارات الإيقاف](https://inference-docs.cerebras.ai/support/deprecation) للحصول على معرّفات endpoints العامة الحالية.
**ملاحظة:** يستخدم هذا المزود LiteLLM. أضفه كتبعية لمشروعك:
```bash
uv add 'crewai[litellm]'
@@ -1189,7 +1011,7 @@ mode: "wide"
# Create an LLM with streaming enabled
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
stream=True # Enable streaming
)
```
@@ -1239,7 +1061,7 @@ mode: "wide"
my_listener = MyCustomListener()
llm = LLM(model="gpt-4o-mini", temperature=0, stream=True)
llm = LLM(model="openai/gpt-5.6-terra", stream=True)
researcher = Agent(
role="About User",
@@ -1319,6 +1141,8 @@ mode: "wide"
يدعم CrewAI الاستجابات المهيكلة من استدعاءات LLM من خلال السماح لك بتحديد `response_format` باستخدام نموذج Pydantic. يمكّن هذا الإطار من تحليل المخرجات والتحقق منها تلقائيًا، مما يسهّل دمج الاستجابة في تطبيقك دون معالجة لاحقة يدوية.
يختلف دعم المخرجات المهيكلة باختلاف المزوّد والنموذج. اختبر النموذج الذي اخترته قبل الاعتماد على الاستجابات المهيكلة في بيئة الإنتاج.
```python Code
from crewai import LLM
@@ -1328,7 +1152,7 @@ class Dog(BaseModel):
breed: str
llm = LLM(model="gpt-4o", response_format=Dog)
llm = LLM(model="openai/gpt-5.6-terra", response_format=Dog)
response = llm.call(
"Analyze the following messages and return the name, age, and breed. "
@@ -1357,8 +1181,8 @@ print(response)
# 3. Task splitting for large contexts
llm = LLM(
model="gpt-4",
max_tokens=4000, # Limit response length
model="openai/gpt-5.6-terra",
max_completion_tokens=4000, # Limit response length
)
```
@@ -1382,15 +1206,14 @@ print(response)
```python
# Configure model with appropriate settings
llm = LLM(
model="openai/gpt-4-turbo-preview",
temperature=0.7, # Adjust based on task
max_tokens=4096, # Set based on output needs
timeout=300 # Longer timeout for complex tasks
model="openai/gpt-5.6-terra",
reasoning_effort="medium",
max_completion_tokens=4096,
timeout=300
)
```
<Tip>
- درجة حرارة منخفضة (0.1 إلى 0.3) للاستجابات الواقعية
- درجة حرارة عالية (0.7 إلى 0.9) للمهام الإبداعية
استخدم عناصر التحكم التي يدعمها النموذج المحدد. حسب المزود، قد تكون `temperature` أو مستوى reasoning أو thinking، أو تعليمات prompt تحدد الأسلوب والتباين المطلوبين.
</Tip>
</Step>

400
docs/edge/en/concepts/llms.mdx
View File

@@ -22,7 +22,10 @@ Large Language Models (LLMs) are the core intelligence behind CrewAI agents. The
The context window determines how much text an LLM can process at once. Larger windows (e.g., 128K tokens) allow for more context but may be more expensive and slower.
</Card>
<Card title="Temperature" icon="temperature-three-quarters">
Temperature (0.0 to 1.0) controls response randomness. Lower values (e.g., 0.2) produce more focused, deterministic outputs, while higher values (e.g., 0.8) increase creativity and variability.
Temperature is a sampling control supported by some models. Lower values
generally make sampling more focused, while higher values increase
variability. Some newer reasoning models ignore, deprecate, or reject this
parameter, so check the selected model's documentation before setting it.
</Card>
<Card title="Provider Selection" icon="server">
Each LLM provider (e.g., OpenAI, Anthropic, Google) offers different models with varying capabilities, pricing, and features. Choose based on your needs for accuracy, speed, and cost.
@@ -38,7 +41,7 @@ There are different places in CrewAI code where you can specify the model to use
The simplest way to get started. Set the model in your environment directly, through an `.env` file or in your app code. If you used `crewai create` to bootstrap your project, it will be set already.
```bash .env
MODEL=model-id # e.g. gpt-4o, gemini-2.0-flash, claude-3-sonnet-...
MODEL=provider/model-id # e.g. openai/gpt-5.6-terra
# Be sure to set your API keys here too. See the Provider
# section below.
@@ -57,7 +60,7 @@ There are different places in CrewAI code where you can specify the model to use
goal: Conduct comprehensive research and analysis
backstory: A dedicated research professional with years of experience
verbose: true
llm: provider/model-id # e.g. openai/gpt-4o, google/gemini-2.0-flash, anthropic/claude...
llm: provider/model-id # e.g. anthropic/claude-sonnet-4-6
# (see provider configuration examples below for more)
```
@@ -76,32 +79,27 @@ There are different places in CrewAI code where you can specify the model to use
from crewai import LLM
# Basic configuration
llm = LLM(model="model-id-here") # gpt-4o, gemini-2.0-flash, anthropic/claude...
llm = LLM(model="provider/model-id") # e.g. gemini/gemini-3.6-flash
# Advanced configuration with detailed parameters
llm = LLM(
model="model-id-here", # gpt-4o, gemini-2.0-flash, anthropic/claude...
temperature=0.7, # Higher for more creative outputs
timeout=120, # Seconds to wait for response
max_tokens=4000, # Maximum length of response
top_p=0.9, # Nucleus sampling parameter
frequency_penalty=0.1 , # Reduce repetition
presence_penalty=0.1, # Encourage topic diversity
model="provider/model-id",
timeout=120,
max_tokens=4000,
response_format={"type": "json"}, # For structured outputs
seed=42 # For reproducible results
)
```
<Info>
Parameter explanations:
- `temperature`: Controls randomness (0.0-1.0)
- `timeout`: Maximum wait time for response
- `max_tokens`: Limits response length
- `top_p`: Alternative to temperature for sampling
- `frequency_penalty`: Reduces word repetition
- `presence_penalty`: Encourages new topics
- `response_format`: Specifies output structure
- `seed`: Ensures consistent outputs
Sampling controls such as `temperature` and `top_p`, penalty parameters,
token-limit names, and reasoning controls are model-specific. Add them
only when the selected provider and model support them. See the provider
examples below and the provider's model documentation.
</Info>
</Tab>
</Tabs>
@@ -120,6 +118,13 @@ There are different places in CrewAI code where you can specify the model to use
CrewAI supports a multitude of LLM providers, each offering unique features, authentication methods, and model capabilities.
In this section, you'll find detailed examples that help you select, configure, and optimize the LLM that best fits your project's needs.
<Warning>
Model availability changes frequently and can vary by account, region, and
cloud platform. The examples below use models that are current at the time of
writing, but they are not exhaustive support lists. Before deploying, verify
the model ID and lifecycle status in the provider's linked model catalog.
</Warning>
<AccordionGroup>
<Accordion title="OpenAI">
CrewAI provides native integration with OpenAI through the OpenAI Python SDK.
@@ -137,10 +142,10 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
api_key="your-api-key", # Or set OPENAI_API_KEY
temperature=0.7,
max_tokens=4000
reasoning_effort="medium",
max_completion_tokens=4000
)
```
@@ -161,25 +166,16 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
api_key="your-api-key",
base_url="https://api.openai.com/v1", # Optional custom endpoint
organization="org-...", # Optional organization ID
project="proj_...", # Optional project ID
temperature=0.7,
max_tokens=4000,
max_completion_tokens=4000, # For newer models
top_p=0.9,
frequency_penalty=0.1,
presence_penalty=0.1,
stop=["END"],
seed=42, # For reproducible outputs
max_completion_tokens=4000,
reasoning_effort="medium",
stream=True, # Enable streaming
timeout=60.0, # Request timeout in seconds
max_retries=3, # Maximum retry attempts
logprobs=True, # Return log probabilities
top_logprobs=5, # Number of most likely tokens
reasoning_effort="medium" # For o1 models: low, medium, high
max_retries=3 # Maximum retry attempts
)
```
@@ -194,7 +190,7 @@ In this section, you'll find detailed examples that help you select, configure,
summary: str
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
)
```
@@ -203,30 +199,18 @@ In this section, you'll find detailed examples that help you select, configure,
- `OPENAI_BASE_URL`: Custom base URL for OpenAI API (optional)
**Features:**
- Native function calling support (except o1 models)
- Native function calling support
- Structured outputs with JSON schema
- Streaming support for real-time responses
- Token usage tracking
- Stop sequences support (except o1 models)
- Provider-specific generation controls
- Log probabilities for token-level insights
- Reasoning effort control for o1 models
- Reasoning effort control for supported models
**Supported Models:**
| Model | Context Window | Best For |
|---------------------|------------------|-----------------------------------------------|
| gpt-4.1 | 1M tokens | Latest model with enhanced capabilities |
| gpt-4.1-mini | 1M tokens | Efficient version with large context |
| gpt-4.1-nano | 1M tokens | Ultra-efficient variant |
| gpt-4o | 128,000 tokens | Optimized for speed and intelligence |
| gpt-4o-mini | 200,000 tokens | Cost-effective with large context |
| gpt-4-turbo | 128,000 tokens | Long-form content, document analysis |
| gpt-4 | 8,192 tokens | High-accuracy tasks, complex reasoning |
| o1 | 200,000 tokens | Advanced reasoning, complex problem-solving |
| o1-preview | 128,000 tokens | Preview of reasoning capabilities |
| o1-mini | 128,000 tokens | Efficient reasoning model |
| o3-mini | 200,000 tokens | Lightweight reasoning model |
| o4-mini | 200,000 tokens | Next-gen efficient reasoning |
OpenAI regularly adds models and retires older snapshots. See the
[OpenAI model catalog](https://developers.openai.com/api/docs/models) for
current model IDs, context windows, endpoint compatibility, and lifecycle
information.
**Responses API:**
@@ -288,14 +272,8 @@ In this section, you'll find detailed examples that help you select, configure,
)
```
All models listed here https://llama.developer.meta.com/docs/models/ are supported.
| Model ID | Input context length | Output context length | Input Modalities | Output Modalities |
| --- | --- | --- | --- | --- |
| `meta_llama/Llama-4-Scout-17B-16E-Instruct-FP8` | 128k | 4028 | Text, Image | Text |
| `meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8` | 128k | 4028 | Text, Image | Text |
| `meta_llama/Llama-3.3-70B-Instruct` | 128k | 4028 | Text | Text |
| `meta_llama/Llama-3.3-8B-Instruct` | 128k | 4028 | Text | Text |
See the [Meta Llama model overview](https://ai.meta.com/llama/get-started/)
for current model families, modalities, and context guidance.
**Note:** This provider uses LiteLLM. Add it as a dependency to your project:
```bash
@@ -365,7 +343,7 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="anthropic/claude-3-5-sonnet-20241022",
model="anthropic/claude-sonnet-4-6",
api_key="your-api-key", # Or set ANTHROPIC_API_KEY
max_tokens=4096 # Required for Anthropic
)
@@ -376,12 +354,10 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="anthropic/claude-3-5-sonnet-20241022",
model="anthropic/claude-sonnet-4-6",
api_key="your-api-key",
base_url="https://api.anthropic.com", # Optional custom endpoint
temperature=0.7,
max_tokens=4096, # Required parameter
top_p=0.9,
stop_sequences=["END", "STOP"], # Anthropic uses stop_sequences
stream=True, # Enable streaming
timeout=60.0, # Request timeout in seconds
@@ -389,7 +365,7 @@ In this section, you'll find detailed examples that help you select, configure,
)
```
**Extended Thinking (Claude Sonnet 4 and Beyond):**
**Extended Thinking:**
CrewAI supports Anthropic's Extended Thinking feature, which allows Claude to think through problems in a more human-like way before responding. This is particularly useful for complex reasoning, analysis, and problem-solving tasks.
@@ -398,14 +374,14 @@ In this section, you'll find detailed examples that help you select, configure,
# Enable extended thinking with default settings
llm = LLM(
model="anthropic/claude-sonnet-4",
model="anthropic/claude-sonnet-4-6",
thinking={"type": "enabled"},
max_tokens=10000
)
# Configure thinking with budget control
llm = LLM(
model="anthropic/claude-sonnet-4",
model="anthropic/claude-sonnet-4-6",
thinking={
"type": "enabled",
"budget_tokens": 5000 # Limit thinking tokens
@@ -418,9 +394,8 @@ In this section, you'll find detailed examples that help you select, configure,
- `type`: Set to `"enabled"` to activate extended thinking mode
- `budget_tokens` (optional): Maximum tokens to use for thinking (helps control costs)
**Models Supporting Extended Thinking:**
- `claude-sonnet-4` and newer models
- `claude-3-7-sonnet` (with extended thinking capabilities)
Thinking modes and accepted parameters vary across Claude generations.
Check the selected model's capabilities before enabling `thinking`.
**When to Use Extended Thinking:**
- Complex reasoning and multi-step problem solving
@@ -436,7 +411,7 @@ In this section, you'll find detailed examples that help you select, configure,
**Features:**
- Native tool use support for Claude 3+ models
- Extended Thinking support for Claude Sonnet 4+
- Extended Thinking support for compatible Claude models
- Streaming support for real-time responses
- Automatic system message handling
- Stop sequences for controlled output
@@ -450,20 +425,10 @@ In this section, you'll find detailed examples that help you select, configure,
- First message must be from the user (automatically handled)
- Messages must alternate between user and assistant
**Supported Models:**
| Model | Context Window | Best For |
|------------------------------|----------------|-----------------------------------------------|
| claude-sonnet-4 | 200,000 tokens | Latest with extended thinking capabilities |
| claude-3-7-sonnet | 200,000 tokens | Advanced reasoning and agentic tasks |
| claude-3-5-sonnet-20241022 | 200,000 tokens | Latest Sonnet with best performance |
| claude-3-5-haiku | 200,000 tokens | Fast, compact model for quick responses |
| claude-3-opus | 200,000 tokens | Most capable for complex tasks |
| claude-3-sonnet | 200,000 tokens | Balanced intelligence and speed |
| claude-3-haiku | 200,000 tokens | Fastest for simple tasks |
| claude-2.1 | 200,000 tokens | Extended context, reduced hallucinations |
| claude-2 | 100,000 tokens | Versatile model for various tasks |
| claude-instant | 100,000 tokens | Fast, cost-effective for everyday tasks |
See Anthropic's [models overview](https://platform.claude.com/docs/en/about-claude/models/overview)
for current model IDs and capabilities, and review the
[model deprecation table](https://platform.claude.com/docs/en/about-claude/model-deprecations)
before pinning a model in production.
**Note:** To use Anthropic, install the required dependencies:
```bash
@@ -495,9 +460,8 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.0-flash",
model="gemini/gemini-3.6-flash",
api_key="your-api-key", # Or set GOOGLE_API_KEY/GEMINI_API_KEY
temperature=0.7
)
```
@@ -506,11 +470,8 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.5-flash",
model="gemini/gemini-3.6-flash",
api_key="your-api-key",
temperature=0.7,
top_p=0.9,
top_k=40, # Top-k sampling parameter
max_output_tokens=8192,
stop_sequences=["END", "STOP"],
stream=True, # Enable streaming
@@ -536,8 +497,7 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.0-flash",
temperature=0.7
model="gemini/gemini-3.6-flash"
)
```
@@ -554,7 +514,7 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="gemini/gemini-1.5-pro",
model="gemini/gemini-3.6-flash",
project="your-gcp-project-id",
location="us-central1" # GCP region
)
@@ -567,7 +527,7 @@ In this section, you'll find detailed examples that help you select, configure,
- `GOOGLE_CLOUD_LOCATION`: GCP location (defaults to `us-central1`)
**Features:**
- Native function calling support for Gemini 1.5+ and 2.x models
- Native function calling support for compatible Gemini models
- Streaming support for real-time responses
- Multimodal capabilities (text, images, video)
- Safety settings configuration
@@ -575,53 +535,24 @@ In this section, you'll find detailed examples that help you select, configure,
- Automatic system instruction handling
- Token usage tracking
**Gemini Models:**
Google offers a range of powerful models optimized for different use cases.
| Model | Context Window | Best For |
|--------------------------------|----------------|-------------------------------------------------------------------|
| gemini-2.5-flash | 1M tokens | Adaptive thinking, cost efficiency |
| gemini-2.5-pro | 1M tokens | Enhanced thinking and reasoning, multimodal understanding |
| gemini-2.0-flash | 1M tokens | Next generation features, speed, thinking |
| gemini-2.0-flash-thinking | 32,768 tokens | Advanced reasoning with thinking process |
| gemini-2.0-flash-lite | 1M tokens | Cost efficiency and low latency |
| gemini-1.5-pro | 2M tokens | Best performing, logical reasoning, coding |
| gemini-1.5-flash | 1M tokens | Balanced multimodal model, good for most tasks |
| gemini-1.5-flash-8b | 1M tokens | Fastest, most cost-efficient |
| gemini-1.0-pro | 32,768 tokens | Earlier generation model |
**Gemma Models:**
The Gemini API also supports [Gemma models](https://ai.google.dev/gemma/docs) hosted on Google infrastructure.
| Model | Context Window | Best For |
|----------------|----------------|------------------------------------|
| gemma-3-1b | 32,000 tokens | Ultra-lightweight tasks |
| gemma-3-4b | 128,000 tokens | Efficient general-purpose tasks |
| gemma-3-12b | 128,000 tokens | Balanced performance and efficiency|
| gemma-3-27b | 128,000 tokens | High-performance tasks |
Google publishes current Gemini IDs, capabilities, and lifecycle stages in
the [Gemini model catalog](https://ai.google.dev/gemini-api/docs/models).
Check the [deprecation schedule](https://ai.google.dev/gemini-api/docs/deprecations)
before choosing a stable or preview model. The Gemini API also hosts
[Gemma models](https://ai.google.dev/gemma/docs).
**Note:** To use Google Gemini, install the required dependencies:
```bash
uv add "crewai[google-genai]"
```
The full list of models is available in the [Gemini model docs](https://ai.google.dev/gemini-api/docs/models).
</Accordion>
<Accordion title="Google (Vertex AI)">
Get credentials from your Google Cloud Console and save it to a JSON file, then load it with the following code:
```python Code
import json
file_path = 'path/to/vertex_ai_service_account.json'
# Load the JSON file
with open(file_path, 'r') as file:
vertex_credentials = json.load(file)
# Convert the credentials to a JSON string
vertex_credentials_json = json.dumps(vertex_credentials)
Authenticate with [Application Default Credentials](https://cloud.google.com/docs/authentication/provide-credentials-adc), then configure the native Gemini provider for Vertex AI:
```toml .env
GOOGLE_GENAI_USE_VERTEXAI=true
GOOGLE_CLOUD_PROJECT=<your-project-id>
GOOGLE_CLOUD_LOCATION=<location>
```
Example usage in your CrewAI project:
@@ -629,27 +560,17 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="gemini-1.5-pro-latest", # or vertex_ai/gemini-1.5-pro-latest
temperature=0.7,
vertex_credentials=vertex_credentials_json
model="gemini/gemini-3.6-flash"
)
```
Google offers a range of powerful models optimized for different use cases:
Vertex AI availability varies by region. Use the
[Vertex AI model catalog](https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models)
to verify the model ID and location before deployment.
| Model | Context Window | Best For |
|--------------------------------|----------------|-------------------------------------------------------------------|
| gemini-2.5-flash-preview-04-17 | 1M tokens | Adaptive thinking, cost efficiency |
| gemini-2.5-pro-preview-05-06 | 1M tokens | Enhanced thinking and reasoning, multimodal understanding, advanced coding, and more |
| gemini-2.0-flash | 1M tokens | Next generation features, speed, thinking, and realtime streaming |
| gemini-2.0-flash-lite | 1M tokens | Cost efficiency and low latency |
| gemini-1.5-flash | 1M tokens | Balanced multimodal model, good for most tasks |
| gemini-1.5-flash-8B | 1M tokens | Fastest, most cost-efficient, good for high-frequency tasks |
| gemini-1.5-pro | 2M tokens | Best performing, wide variety of reasoning tasks including logical reasoning, coding, and creative collaboration |
**Note:** This provider uses LiteLLM. Add it as a dependency to your project:
**Note:** This route uses CrewAI's native Gemini integration. Add it as a dependency to your project:
```bash
uv add 'crewai[litellm]'
uv add "crewai[google-genai]"
```
</Accordion>
@@ -740,7 +661,7 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
model="bedrock/us.anthropic.claude-sonnet-4-6",
region_name="us-east-1"
)
```
@@ -750,7 +671,7 @@ In this section, you'll find detailed examples that help you select, configure,
from crewai import LLM
llm = LLM(
model="bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
model="bedrock/us.anthropic.claude-sonnet-4-6",
aws_access_key_id="your-access-key", # Or set AWS_ACCESS_KEY_ID
aws_secret_access_key="your-secret-key", # Or set AWS_SECRET_ACCESS_KEY
aws_session_token="your-session-token", # For temporary credentials
@@ -795,38 +716,9 @@ In this section, you'll find detailed examples that help you select, configure,
- First message must be from user (automatically handled)
- Some models (like Cohere) require conversation to end with user message
[Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html) is a managed service that provides access to multiple foundation models from top AI companies through a unified API.
| Model | Context Window | Best For |
|-------------------------|----------------------|-------------------------------------------------------------------|
| Amazon Nova Pro | Up to 300k tokens | High-performance, model balancing accuracy, speed, and cost-effectiveness across diverse tasks. |
| Amazon Nova Micro | Up to 128k tokens | High-performance, cost-effective text-only model optimized for lowest latency responses. |
| Amazon Nova Lite | Up to 300k tokens | High-performance, affordable multimodal processing for images, video, and text with real-time capabilities. |
| Claude 3.7 Sonnet | Up to 128k tokens | High-performance, best for complex reasoning, coding & AI agents |
| Claude 3.5 Sonnet v2 | Up to 200k tokens | State-of-the-art model specialized in software engineering, agentic capabilities, and computer interaction at optimized cost. |
| Claude 3.5 Sonnet | Up to 200k tokens | High-performance model delivering superior intelligence and reasoning across diverse tasks with optimal speed-cost balance. |
| Claude 3.5 Haiku | Up to 200k tokens | Fast, compact multimodal model optimized for quick responses and seamless human-like interactions |
| Claude 3 Sonnet | Up to 200k tokens | Multimodal model balancing intelligence and speed for high-volume deployments. |
| Claude 3 Haiku | Up to 200k tokens | Compact, high-speed multimodal model optimized for quick responses and natural conversational interactions |
| Claude 3 Opus | Up to 200k tokens | Most advanced multimodal model exceling at complex tasks with human-like reasoning and superior contextual understanding. |
| Claude 2.1 | Up to 200k tokens | Enhanced version with expanded context window, improved reliability, and reduced hallucinations for long-form and RAG applications |
| Claude | Up to 100k tokens | Versatile model excelling in sophisticated dialogue, creative content, and precise instruction following. |
| Claude Instant | Up to 100k tokens | Fast, cost-effective model for everyday tasks like dialogue, analysis, summarization, and document Q&A |
| Llama 3.1 405B Instruct | Up to 128k tokens | Advanced LLM for synthetic data generation, distillation, and inference for chatbots, coding, and domain-specific tasks. |
| Llama 3.1 70B Instruct | Up to 128k tokens | Powers complex conversations with superior contextual understanding, reasoning and text generation. |
| Llama 3.1 8B Instruct | Up to 128k tokens | Advanced state-of-the-art model with language understanding, superior reasoning, and text generation. |
| Llama 3 70B Instruct | Up to 8k tokens | Powers complex conversations with superior contextual understanding, reasoning and text generation. |
| Llama 3 8B Instruct | Up to 8k tokens | Advanced state-of-the-art LLM with language understanding, superior reasoning, and text generation. |
| Titan Text G1 - Lite | Up to 4k tokens | Lightweight, cost-effective model optimized for English tasks and fine-tuning with focus on summarization and content generation. |
| Titan Text G1 - Express | Up to 8k tokens | Versatile model for general language tasks, chat, and RAG applications with support for English and 100+ languages. |
| Cohere Command | Up to 4k tokens | Model specialized in following user commands and delivering practical enterprise solutions. |
| Jurassic-2 Mid | Up to 8,191 tokens | Cost-effective model balancing quality and affordability for diverse language tasks like Q&A, summarization, and content generation. |
| Jurassic-2 Ultra | Up to 8,191 tokens | Model for advanced text generation and comprehension, excelling in complex tasks like analysis and content creation. |
| Jamba-Instruct | Up to 256k tokens | Model with extended context window optimized for cost-effective text generation, summarization, and Q&A. |
| Mistral 7B Instruct | Up to 32k tokens | This LLM follows instructions, completes requests, and generates creative text. |
| Mistral 8x7B Instruct | Up to 32k tokens | An MOE LLM that follows instructions, completes requests, and generates creative text. |
| DeepSeek R1 | 32,768 tokens | Advanced reasoning model |
Amazon Bedrock model access and IDs vary by region. Use AWS's
[supported models and regions](https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html)
reference to select a model and verify Converse API support.
**Note:** To use AWS Bedrock, install the required dependencies:
```bash
uv add "crewai[bedrock]"
@@ -882,81 +774,14 @@ In this section, you'll find detailed examples that help you select, configure,
Example usage in your CrewAI project:
```python Code
llm = LLM(
model="nvidia_nim/meta/llama3-70b-instruct",
model="nvidia_nim/nvidia/nvidia-nemotron-3-ultra-550b-a55b",
temperature=0.7
)
```
Nvidia NIM provides a comprehensive suite of models for various use cases, from general-purpose tasks to specialized applications.
| Model | Context Window | Best For |
|-------------------------------------------------------------------------|----------------|-------------------------------------------------------------------|
| nvidia/mistral-nemo-minitron-8b-8k-instruct | 8,192 tokens | State-of-the-art small language model delivering superior accuracy for chatbot, virtual assistants, and content generation. |
| nvidia/nemotron-4-mini-hindi-4b-instruct | 4,096 tokens | A bilingual Hindi-English SLM for on-device inference, tailored specifically for Hindi Language. |
| nvidia/llama-3.1-nemotron-70b-instruct | 128k tokens | Customized for enhanced helpfulness in responses |
| nvidia/llama3-chatqa-1.5-8b | 128k tokens | Advanced LLM to generate high-quality, context-aware responses for chatbots and search engines. |
| nvidia/llama3-chatqa-1.5-70b | 128k tokens | Advanced LLM to generate high-quality, context-aware responses for chatbots and search engines. |
| nvidia/vila | 128k tokens | Multi-modal vision-language model that understands text/img/video and creates informative responses |
| nvidia/neva-22 | 4,096 tokens | Multi-modal vision-language model that understands text/images and generates informative responses |
| nvidia/nemotron-mini-4b-instruct | 8,192 tokens | General-purpose tasks |
| nvidia/usdcode-llama3-70b-instruct | 128k tokens | State-of-the-art LLM that answers OpenUSD knowledge queries and generates USD-Python code. |
| nvidia/nemotron-4-340b-instruct | 4,096 tokens | Creates diverse synthetic data that mimics the characteristics of real-world data. |
| meta/codellama-70b | 100k tokens | LLM capable of generating code from natural language and vice versa. |
| meta/llama2-70b | 4,096 tokens | Cutting-edge large language AI model capable of generating text and code in response to prompts. |
| meta/llama3-8b-instruct | 8,192 tokens | Advanced state-of-the-art LLM with language understanding, superior reasoning, and text generation. |
| meta/llama3-70b-instruct | 8,192 tokens | Powers complex conversations with superior contextual understanding, reasoning and text generation. |
| meta/llama-3.1-8b-instruct | 128k tokens | Advanced state-of-the-art model with language understanding, superior reasoning, and text generation. |
| meta/llama-3.1-70b-instruct | 128k tokens | Powers complex conversations with superior contextual understanding, reasoning and text generation. |
| meta/llama-3.1-405b-instruct | 128k tokens | Advanced LLM for synthetic data generation, distillation, and inference for chatbots, coding, and domain-specific tasks. |
| meta/llama-3.2-1b-instruct | 128k tokens | Advanced state-of-the-art small language model with language understanding, superior reasoning, and text generation. |
| meta/llama-3.2-3b-instruct | 128k tokens | Advanced state-of-the-art small language model with language understanding, superior reasoning, and text generation. |
| meta/llama-3.2-11b-vision-instruct | 128k tokens | Advanced state-of-the-art small language model with language understanding, superior reasoning, and text generation. |
| meta/llama-3.2-90b-vision-instruct | 128k tokens | Advanced state-of-the-art small language model with language understanding, superior reasoning, and text generation. |
| google/gemma-7b | 8,192 tokens | Cutting-edge text generation model text understanding, transformation, and code generation. |
| google/gemma-2b | 8,192 tokens | Cutting-edge text generation model text understanding, transformation, and code generation. |
| google/codegemma-7b | 8,192 tokens | Cutting-edge model built on Google's Gemma-7B specialized for code generation and code completion. |
| google/codegemma-1.1-7b | 8,192 tokens | Advanced programming model for code generation, completion, reasoning, and instruction following. |
| google/recurrentgemma-2b | 8,192 tokens | Novel recurrent architecture based language model for faster inference when generating long sequences. |
| google/gemma-2-9b-it | 8,192 tokens | Cutting-edge text generation model text understanding, transformation, and code generation. |
| google/gemma-2-27b-it | 8,192 tokens | Cutting-edge text generation model text understanding, transformation, and code generation. |
| google/gemma-2-2b-it | 8,192 tokens | Cutting-edge text generation model text understanding, transformation, and code generation. |
| google/deplot | 512 tokens | One-shot visual language understanding model that translates images of plots into tables. |
| google/paligemma | 8,192 tokens | Vision language model adept at comprehending text and visual inputs to produce informative responses. |
| mistralai/mistral-7b-instruct-v0.2 | 32k tokens | This LLM follows instructions, completes requests, and generates creative text. |
| mistralai/mixtral-8x7b-instruct-v0.1 | 8,192 tokens | An MOE LLM that follows instructions, completes requests, and generates creative text. |
| mistralai/mistral-large | 4,096 tokens | Creates diverse synthetic data that mimics the characteristics of real-world data. |
| mistralai/mixtral-8x22b-instruct-v0.1 | 8,192 tokens | Creates diverse synthetic data that mimics the characteristics of real-world data. |
| mistralai/mistral-7b-instruct-v0.3 | 32k tokens | This LLM follows instructions, completes requests, and generates creative text. |
| nv-mistralai/mistral-nemo-12b-instruct | 128k tokens | Most advanced language model for reasoning, code, multilingual tasks; runs on a single GPU. |
| mistralai/mamba-codestral-7b-v0.1 | 256k tokens | Model for writing and interacting with code across a wide range of programming languages and tasks. |
| microsoft/phi-3-mini-128k-instruct | 128K tokens | Lightweight, state-of-the-art open LLM with strong math and logical reasoning skills. |
| microsoft/phi-3-mini-4k-instruct | 4,096 tokens | Lightweight, state-of-the-art open LLM with strong math and logical reasoning skills. |
| microsoft/phi-3-small-8k-instruct | 8,192 tokens | Lightweight, state-of-the-art open LLM with strong math and logical reasoning skills. |
| microsoft/phi-3-small-128k-instruct | 128K tokens | Lightweight, state-of-the-art open LLM with strong math and logical reasoning skills. |
| microsoft/phi-3-medium-4k-instruct | 4,096 tokens | Lightweight, state-of-the-art open LLM with strong math and logical reasoning skills. |
| microsoft/phi-3-medium-128k-instruct | 128K tokens | Lightweight, state-of-the-art open LLM with strong math and logical reasoning skills. |
| microsoft/phi-3.5-mini-instruct | 128K tokens | Lightweight multilingual LLM powering AI applications in latency bound, memory/compute constrained environments |
| microsoft/phi-3.5-moe-instruct | 128K tokens | Advanced LLM based on Mixture of Experts architecture to deliver compute efficient content generation |
| microsoft/kosmos-2 | 1,024 tokens | Groundbreaking multimodal model designed to understand and reason about visual elements in images. |
| microsoft/phi-3-vision-128k-instruct | 128k tokens | Cutting-edge open multimodal model exceling in high-quality reasoning from images. |
| microsoft/phi-3.5-vision-instruct | 128k tokens | Cutting-edge open multimodal model exceling in high-quality reasoning from images. |
| databricks/dbrx-instruct | 12k tokens | A general-purpose LLM with state-of-the-art performance in language understanding, coding, and RAG. |
| snowflake/arctic | 1,024 tokens | Delivers high efficiency inference for enterprise applications focused on SQL generation and coding. |
| aisingapore/sea-lion-7b-instruct | 4,096 tokens | LLM to represent and serve the linguistic and cultural diversity of Southeast Asia |
| ibm/granite-8b-code-instruct | 4,096 tokens | Software programming LLM for code generation, completion, explanation, and multi-turn conversion. |
| ibm/granite-34b-code-instruct | 8,192 tokens | Software programming LLM for code generation, completion, explanation, and multi-turn conversion. |
| ibm/granite-3.0-8b-instruct | 4,096 tokens | Advanced Small Language Model supporting RAG, summarization, classification, code, and agentic AI |
| ibm/granite-3.0-3b-a800m-instruct | 4,096 tokens | Highly efficient Mixture of Experts model for RAG, summarization, entity extraction, and classification |
| mediatek/breeze-7b-instruct | 4,096 tokens | Creates diverse synthetic data that mimics the characteristics of real-world data. |
| upstage/solar-10.7b-instruct | 4,096 tokens | Excels in NLP tasks, particularly in instruction-following, reasoning, and mathematics. |
| writer/palmyra-med-70b-32k | 32k tokens | Leading LLM for accurate, contextually relevant responses in the medical domain. |
| writer/palmyra-med-70b | 32k tokens | Leading LLM for accurate, contextually relevant responses in the medical domain. |
| writer/palmyra-fin-70b-32k | 32k tokens | Specialized LLM for financial analysis, reporting, and data processing |
| 01-ai/yi-large | 32k tokens | Powerful model trained on English and Chinese for diverse tasks including chatbot and creative writing. |
| deepseek-ai/deepseek-coder-6.7b-instruct | 2k tokens | Powerful coding model offering advanced capabilities in code generation, completion, and infilling |
| rakuten/rakutenai-7b-instruct | 1,024 tokens | Advanced state-of-the-art LLM with language understanding, superior reasoning, and text generation. |
| rakuten/rakutenai-7b-chat | 1,024 tokens | Advanced state-of-the-art LLM with language understanding, superior reasoning, and text generation. |
| baichuan-inc/baichuan2-13b-chat | 4,096 tokens | Support Chinese and English chat, coding, math, instruction following, solving quizzes |
NVIDIA NIM's hosted catalog changes frequently. Use the
[NVIDIA NIM model catalog](https://build.nvidia.com/models) to select a
current endpoint and verify its model ID, modalities, and context limits.
**Note:** This provider uses LiteLLM. Add it as a dependency to your project:
```bash
@@ -1074,15 +899,15 @@ In this section, you'll find detailed examples that help you select, configure,
Example usage in your CrewAI project:
```python Code
llm = LLM(
model="groq/llama-3.2-90b-text-preview",
model="groq/qwen/qwen3.6-27b",
temperature=0.7
)
```
| Model | Context Window | Best For |
|-------------------|------------------|--------------------------------------------|
| Llama 3.1 70B/8B | 131,072 tokens | High-performance, large context tasks |
| Llama 3.2 Series | 8,192 tokens | General-purpose tasks |
| Mixtral 8x7B | 32,768 tokens | Balanced performance and context |
Groq distinguishes production and preview models and retires model IDs
regularly. Check the [Groq model catalog](https://console.groq.com/docs/models)
and [deprecation page](https://console.groq.com/docs/deprecations) before
selecting a model for production.
**Note:** This provider uses LiteLLM. Add it as a dependency to your project:
```bash
@@ -1164,11 +989,14 @@ In this section, you'll find detailed examples that help you select, configure,
Example usage in your CrewAI project:
```python Code
llm = LLM(
model="llama-3.1-sonar-large-128k-online",
base_url="https://api.perplexity.ai/"
model="perplexity/sonar-pro"
)
```
See the [Perplexity model catalog](https://docs.perplexity.ai/getting-started/models)
and [changelog](https://docs.perplexity.ai/docs/resources/changelog) for
current model IDs and deprecation notices.
**Note:** This provider uses LiteLLM. Add it as a dependency to your project:
```bash
uv add 'crewai[litellm]'
@@ -1204,17 +1032,14 @@ In this section, you'll find detailed examples that help you select, configure,
Example usage in your CrewAI project:
```python Code
llm = LLM(
model="sambanova/Meta-Llama-3.1-8B-Instruct",
model="sambanova/Meta-Llama-3.3-70B-Instruct",
temperature=0.7
)
```
| Model | Context Window | Best For |
|--------------------|------------------------|----------------------------------------------|
| Llama 3.1 70B/8B | Up to 131,072 tokens | High-performance, large context tasks |
| Llama 3.1 405B | 8,192 tokens | High-performance and output quality |
| Llama 3.2 Series | 8,192 tokens | General-purpose, multimodal tasks |
| Llama 3.3 70B | Up to 131,072 tokens | High-performance and output quality |
| Qwen2 familly | 8,192 tokens | High-performance and output quality |
SambaNova Cloud's hosted models can change independently of CrewAI. Query
the [models endpoint](https://docs.sambanova.ai/docs/api-reference/models/get-environments-available-model-list-metadata)
and check the [deprecation guide](https://docs.sambanova.ai/docs/en/models/deprecations)
before deployment.
**Note:** This provider uses LiteLLM. Add it as a dependency to your project:
```bash
@@ -1232,7 +1057,7 @@ In this section, you'll find detailed examples that help you select, configure,
Example usage in your CrewAI project:
```python Code
llm = LLM(
model="cerebras/llama3.1-70b",
model="cerebras/gpt-oss-120b",
temperature=0.7,
max_tokens=8192
)
@@ -1246,6 +1071,10 @@ In this section, you'll find detailed examples that help you select, configure,
- Support for long context windows
</Info>
See the [Cerebras model catalog](https://inference-docs.cerebras.ai/models/overview)
and [deprecation notices](https://inference-docs.cerebras.ai/support/deprecation)
for current public endpoint IDs.
**Note:** This provider uses LiteLLM. Add it as a dependency to your project:
```bash
uv add 'crewai[litellm]'
@@ -1320,7 +1149,7 @@ CrewAI supports streaming responses from LLMs, allowing your application to rece
# Create an LLM with streaming enabled
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
stream=True # Enable streaming
)
```
@@ -1370,7 +1199,7 @@ CrewAI supports streaming responses from LLMs, allowing your application to rece
my_listener = MyCustomListener()
llm = LLM(model="gpt-4o-mini", temperature=0, stream=True)
llm = LLM(model="openai/gpt-5.6-terra", stream=True)
researcher = Agent(
role="About User",
@@ -1450,6 +1279,8 @@ CrewAI supports asynchronous LLM calls for improved performance and concurrency
CrewAI supports structured responses from LLM calls by allowing you to define a `response_format` using a Pydantic model. This enables the framework to automatically parse and validate the output, making it easier to integrate the response into your application without manual post-processing.
Structured output support varies by provider and model. Test your chosen model before relying on structured responses in production.
For example, you can define a Pydantic model to represent the expected response structure and pass it as the `response_format` when instantiating the LLM. The model will then be used to convert the LLM output into a structured Python object.
```python Code
@@ -1461,7 +1292,7 @@ class Dog(BaseModel):
breed: str
llm = LLM(model="gpt-4o", response_format=Dog)
llm = LLM(model="openai/gpt-5.6-terra", response_format=Dog)
response = llm.call(
"Analyze the following messages and return the name, age, and breed. "
@@ -1490,8 +1321,8 @@ Learn how to get the most out of your LLM configuration:
# 3. Task splitting for large contexts
llm = LLM(
model="gpt-4",
max_tokens=4000, # Limit response length
model="openai/gpt-5.6-terra",
max_completion_tokens=4000, # Limit response length
)
```
@@ -1515,15 +1346,16 @@ Learn how to get the most out of your LLM configuration:
```python
# Configure model with appropriate settings
llm = LLM(
model="openai/gpt-4-turbo-preview",
temperature=0.7, # Adjust based on task
max_tokens=4096, # Set based on output needs
timeout=300 # Longer timeout for complex tasks
model="openai/gpt-5.6-terra",
reasoning_effort="medium",
max_completion_tokens=4096,
timeout=300
)
```
<Tip>
- Lower temperature (0.1 to 0.3) for factual responses
- Higher temperature (0.7 to 0.9) for creative tasks
Use the controls supported by your selected model. Depending on the
provider, this may be `temperature`, a reasoning or thinking level,
or prompt instructions that define the desired style and variability.
</Tip>
</Step>

292
docs/edge/ko/concepts/llms.mdx
View File

@@ -21,7 +21,7 @@ Large Language Models(LLM)는 CrewAI 에이전트의 핵심 지능입니다. 에
컨텍스트 윈도우는 LLM이 한 번에 처리할 수 있는 텍스트 양을 결정합니다. 더 큰 윈도우(예: 128K 토큰)는 더 많은 문맥을 다룰 수 있지만, 비용과 속도 면에서 더 부담이 될 수 있습니다.
</Card>
<Card title="Temperature" icon="temperature-three-quarters">
Temperature(0.0에서 1.0)는 응답의 무작위성을 조절합니다. 낮은 값(예: 0.2)은 더 집중적이고 결정적인 결과를, 높은 값(예: 0.8)은 창의성과 다양성을 높입니다.
Temperature는 일부 모델이 지원하는 샘플링 제어 옵션입니다. 값이 낮을수록 일반적으로 샘플링이 더 집중되고, 값이 높을수록 변동성이 커집니다. 일부 최신 추론 모델은 이 파라미터를 무시하거나 더 이상 권장하지 않거나 거부하므로, 설정하기 전에 선택한 모델의 문서를 확인하세요.
</Card>
<Card title="제공자 선택" icon="server">
각 LLM 제공자(예: OpenAI, Anthropic, Google)는 다양한 기능, 가격, 특성을 가진 모델을 제공합니다. 정확성, 속도, 비용 등 요구 사항에 따라 선택하세요.
@@ -37,7 +37,7 @@ CrewAI 코드 내에는 사용할 모델을 지정할 수 있는 여러 위치
가장 간단하게 시작할 수 있는 방법입니다. `.env` 파일이나 앱 코드에서 환경 변수로 직접 모델을 설정할 수 있습니다. `crewai create`를 사용해 프로젝트를 부트스트랩했다면 이미 설정되어 있을 수 있습니다.
```bash .env
MODEL=model-id # e.g. gpt-4o, gemini-2.0-flash, claude-3-sonnet-...
MODEL=provider/model-id # e.g. openai/gpt-5.6-terra
# 반드시 여기에서 API 키도 설정하세요. 아래 제공자
# 섹션을 참고하세요.
@@ -56,7 +56,7 @@ CrewAI 코드 내에는 사용할 모델을 지정할 수 있는 여러 위치
goal: Conduct comprehensive research and analysis
backstory: A dedicated research professional with years of experience
verbose: true
llm: provider/model-id # e.g. openai/gpt-4o, google/gemini-2.0-flash, anthropic/claude...
llm: provider/model-id # e.g. anthropic/claude-sonnet-4-6
# (아래 제공자 구성 예제 참고)
```
@@ -75,32 +75,24 @@ CrewAI 코드 내에는 사용할 모델을 지정할 수 있는 여러 위치
from crewai import LLM
# 기본 설정
llm = LLM(model="model-id-here") # gpt-4o, gemini-2.0-flash, anthropic/claude...
llm = LLM(model="provider/model-id") # e.g. gemini/gemini-3.6-flash
# 자세한 파라미터로 고급 설정
llm = LLM(
model="model-id-here", # gpt-4o, gemini-2.0-flash, anthropic/claude...
temperature=0.7, # 더욱 창의적인 결과를 원할 때 높게 설정
timeout=120, # 응답을 기다릴 최대 초
max_tokens=4000, # 응답의 최대 길이
top_p=0.9, # 누클리어스 샘플링 파라미터
frequency_penalty=0.1 , # 반복 줄이기
presence_penalty=0.1, # 주제 다양성 높이기
model="provider/model-id",
timeout=120,
max_tokens=4000,
response_format={"type": "json"}, # 구조화된 출력용
seed=42 # 결과 재현성 확보용
)
```
<Info>
파라미터 설명:
- `temperature`: 랜덤성 제어 (0.0-1.0)
- `timeout`: 응답 대기 최대 시간
- `max_tokens`: 응답 길이 제한
- `top_p`: 샘플링 시 temperature의 대체값
- `frequency_penalty`: 단어 반복 감소
- `presence_penalty`: 새로운 주제 생성 유도
- `response_format`: 출력 구조 지정
- `seed`: 일관된 출력 보장
`temperature`, `top_p` 같은 샘플링 제어, 페널티 파라미터, 토큰 제한 파라미터 이름, 추론 제어는 모델별로 다릅니다. 선택한 제공자와 모델이 지원하는 경우에만 추가하세요. 아래 제공자 예시와 해당 제공자의 모델 문서를 참고하세요.
</Info>
</Tab>
</Tabs>
@@ -119,6 +111,10 @@ CrewAI 코드 내에는 사용할 모델을 지정할 수 있는 여러 위치
CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양한 LLM 공급자를 지원합니다.
이 섹션에서는 프로젝트의 요구에 가장 적합한 LLM을 선택, 구성, 최적화하는 데 도움이 되는 자세한 예시를 제공합니다.
<Warning>
모델 가용성은 자주 변경되며 계정, 리전, 클라우드 플랫폼에 따라 달라질 수 있습니다. 아래 예시는 작성 시점에 제공되는 모델을 사용하지만 전체 지원 목록은 아닙니다. 배포하기 전에 연결된 제공자 모델 카탈로그에서 모델 ID와 수명 주기 상태를 확인하세요.
</Warning>
<AccordionGroup>
<Accordion title="OpenAI">
`.env` 파일에 다음 환경 변수를 설정하십시오:
@@ -137,28 +133,13 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
from crewai import LLM
llm = LLM(
model="openai/gpt-4", # call model by provider/model_name
temperature=0.8,
max_tokens=150,
top_p=0.9,
frequency_penalty=0.1,
presence_penalty=0.1,
stop=["END"],
seed=42
model="openai/gpt-5.6-terra",
reasoning_effort="medium",
max_completion_tokens=4000
)
```
OpenAI는 다양한 모델과 기능을 제공하는 대표적인 LLM 공급자 중 하나입니다.
| 모델 | 컨텍스트 윈도우 | 최적 용도 |
|-------------------|-------------------|-----------------------------------------------|
| GPT-4 | 8,192 토큰 | 고정확도 작업, 복잡한 추론 |
| GPT-4 Turbo | 128,000 토큰 | 장문 콘텐츠, 문서 분석 |
| GPT-4o & GPT-4o-mini | 128,000 토큰 | 비용 효율적인 대용량 컨텍스트 처리 |
| o3-mini | 200,000 토큰 | 빠른 추론, 복잡한 추론 |
| o1-mini | 128,000 토큰 | 빠른 추론, 복잡한 추론 |
| o1-preview | 128,000 토큰 | 빠른 추론, 복잡한 추론 |
| o1 | 200,000 토큰 | 빠른 추론, 복잡한 추론 |
OpenAI는 정기적으로 모델을 추가하고 이전 스냅샷을 폐기합니다. 현재 모델 ID, 컨텍스트 윈도우, 엔드포인트 호환성, 수명 주기 정보는 [OpenAI 모델 카탈로그](https://developers.openai.com/api/docs/models)를 확인하세요.
**Responses API:**
@@ -215,14 +196,7 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
)
```
https://llama.developer.meta.com/docs/models/ 에 기재된 모든 모델이 지원됩니다.
| 모델 ID | 입력 컨텍스트 길이 | 출력 컨텍스트 길이 | 입력 모달리티 | 출력 모달리티 |
| ------- | ------------------ | ------------------ | ---------------- | ---------------- |
| `meta_llama/Llama-4-Scout-17B-16E-Instruct-FP8` | 128k | 4028 | 텍스트, 이미지 | 텍스트 |
| `meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8` | 128k | 4028 | 텍스트, 이미지 | 텍스트 |
| `meta_llama/Llama-3.3-70B-Instruct` | 128k | 4028 | 텍스트 | 텍스트 |
| `meta_llama/Llama-3.3-8B-Instruct` | 128k | 4028 | 텍스트 | 텍스트 |
현재 모델 제품군, 모달리티, 컨텍스트 지침은 [Meta Llama 모델 개요](https://ai.meta.com/llama/get-started/)를 확인하세요.
**참고:** 이 제공자는 LiteLLM을 사용합니다. 프로젝트에 의존성으로 추가하세요:
```bash
@@ -291,10 +265,12 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
CrewAI 프로젝트에서의 예시 사용법:
```python Code
llm = LLM(
model="anthropic/claude-3-sonnet-20240229-v1:0",
temperature=0.7
model="anthropic/claude-sonnet-4-6",
max_tokens=4096
)
```
현재 모델 ID와 기능은 Anthropic의 [모델 개요](https://platform.claude.com/docs/en/about-claude/models/overview)를 확인하고, 프로덕션에서 모델을 고정하기 전에 [모델 지원 중단 표](https://platform.claude.com/docs/en/about-claude/model-deprecations)를 검토하세요.
</Accordion>
<Accordion title="Google (Gemini API)">
@@ -319,8 +295,7 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.0-flash",
temperature=0.7,
model="gemini/gemini-3.6-flash",
)
```
@@ -339,8 +314,7 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.0-flash",
temperature=0.7
model="gemini/gemini-3.6-flash"
)
```
@@ -352,47 +326,15 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
자세한 내용은 [Vertex AI Express 모드 문서](https://docs.cloud.google.com/vertex-ai/generative-ai/docs/start/quickstart?usertype=apikey)를 참조하세요.
</Info>
### Gemini 모델
Google은 다양한 용도에 최적화된 강력한 모델을 제공합니다.
| 모델 | 컨텍스트 윈도우 | 최적 용도 |
|----------------------------------|-----------------|------------------------------------------------------------------------|
| gemini-2.5-flash-preview-04-17 | 1M 토큰 | 적응형 사고, 비용 효율성 |
| gemini-2.5-pro-preview-05-06 | 1M 토큰 | 향상된 사고 및 추론, 멀티모달 이해, 고급 코딩 등 |
| gemini-2.0-flash | 1M 토큰 | 차세대 기능, 속도, 사고, 실시간 스트리밍 |
| gemini-2.0-flash-lite | 1M 토큰 | 비용 효율성과 낮은 대기 시간 |
| gemini-1.5-flash | 1M 토큰 | 밸런스 잡힌 멀티모달 모델, 대부분의 작업에 적합 |
| gemini-1.5-flash-8B | 1M 토큰 | 가장 빠르고, 비용 효율적, 고빈도 작업에 적합 |
| gemini-1.5-pro | 2M 토큰 | 최고의 성능, 논리적 추론, 코딩, 창의적 협업 등 다양한 추론 작업에 적합 |
전체 모델 목록은 [Gemini 모델 문서](https://ai.google.dev/gemini-api/docs/models)에서 확인할 수 있습니다.
### Gemma
Gemini API를 통해 Google 인프라에서 호스팅되는 [Gemma 모델](https://ai.google.dev/gemma/docs)도 API 키를 이용해 사용할 수 있습니다.
| 모델 | 컨텍스트 윈도우 |
|----------------|----------------|
| gemma-3-1b-it | 32k 토큰 |
| gemma-3-4b-it | 32k 토큰 |
| gemma-3-12b-it | 32k 토큰 |
| gemma-3-27b-it | 128k 토큰 |
Google은 현재 Gemini ID, 기능, 수명 주기 단계를 [Gemini 모델 카탈로그](https://ai.google.dev/gemini-api/docs/models)에 게시합니다. 안정 또는 preview 모델을 선택하기 전에 [지원 중단 일정](https://ai.google.dev/gemini-api/docs/deprecations)을 확인하세요. Gemini API는 [Gemma 모델](https://ai.google.dev/gemma/docs)도 호스팅합니다.
</Accordion>
<Accordion title="Google (Vertex AI)">
Google Cloud Console에서 자격증명을 받아 JSON 파일로 저장한 후, 다음 코드로 로드하세요:
```python Code
import json
file_path = 'path/to/vertex_ai_service_account.json'
# Load the JSON file
with open(file_path, 'r') as file:
vertex_credentials = json.load(file)
# Convert the credentials to a JSON string
vertex_credentials_json = json.dumps(vertex_credentials)
[애플리케이션 기본 사용자 인증 정보](https://cloud.google.com/docs/authentication/provide-credentials-adc)로 인증한 다음, Vertex AI를 사용하도록 네이티브 Gemini 제공업체를 구성하세요:
```toml .env
GOOGLE_GENAI_USE_VERTEXAI=true
GOOGLE_CLOUD_PROJECT=<your-project-id>
GOOGLE_CLOUD_LOCATION=<location>
```
CrewAI 프로젝트에서의 예시 사용법:
@@ -400,27 +342,15 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
from crewai import LLM
llm = LLM(
model="gemini-1.5-pro-latest", # or vertex_ai/gemini-1.5-pro-latest
temperature=0.7,
vertex_credentials=vertex_credentials_json
model="gemini/gemini-3.6-flash"
)
```
Google은 다양한 용도에 최적화된 강력한 모델들을 제공합니다:
사용 가능한 Vertex AI 모델과 리전은 [Vertex AI 모델 정보](https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models)를 확인하세요.
| 모델 | 컨텍스트 윈도우 | 최적 용도 |
|----------------------------------|-----------------|------------------------------------------------------------------------|
| gemini-2.5-flash-preview-04-17 | 1M 토큰 | 적응형 사고, 비용 효율성 |
| gemini-2.5-pro-preview-05-06 | 1M 토큰 | 향상된 사고 및 추론, 멀티모달 이해, 고급 코딩 등 |
| gemini-2.0-flash | 1M 토큰 | 차세대 기능, 속도, 사고, 실시간 스트리밍 |
| gemini-2.0-flash-lite | 1M 토큰 | 비용 효율성과 낮은 대기 시간 |
| gemini-1.5-flash | 1M 토큰 | 밸런스 잡힌 멀티모달 모델, 대부분의 작업에 적합 |
| gemini-1.5-flash-8B | 1M 토큰 | 가장 빠르고, 비용 효율적, 고빈도 작업에 적합 |
| gemini-1.5-pro | 2M 토큰 | 최고의 성능, 논리적 추론, 코딩, 창의적 협업 등 다양한 추론 작업에 적합 |
**참고:** 이 제공자는 LiteLLM을 사용합니다. 프로젝트에 의존성으로 추가하세요:
**참고:** 이 경로는 CrewAI의 네이티브 Gemini 통합을 사용합니다. 프로젝트에 의존성으로 추가하세요:
```bash
uv add 'crewai[litellm]'
uv add "crewai[google-genai]"
```
</Accordion>
@@ -455,7 +385,7 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
CrewAI 프로젝트에서의 예시 사용법:
```python Code
llm = LLM(
model="bedrock/anthropic.claude-3-sonnet-20240229-v1:0"
model="bedrock/us.anthropic.claude-sonnet-4-6"
)
```
@@ -463,34 +393,6 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
[Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html)은 대표적인 AI 회사들의 여러 파운데이션 모델에 통합 API를 통해 접근할 수 있는 매니지드 서비스로, 안전하고 책임감 있는 AI 응용프로그램 개발을 가능하게 해줍니다.
| 모델 | 컨텍스트 윈도우 | 최적 용도 |
|-----------------------------|--------------------|------------------------------------------------------------------------|
| Amazon Nova Pro | 최대 300k 토큰 | 다양한 작업에서 정확성, 속도, 비용을 균형 있게 제공하는 고성능 모델 |
| Amazon Nova Micro | 최대 128k 토큰 | 텍스트 전용, 최소 레이턴시 응답에 최적화된 비용 효율적 고성능 모델 |
| Amazon Nova Lite | 최대 300k 토큰 | 이미지, 비디오, 텍스트를 아우르는 실시간 멀티모달 처리 |
| Claude 3.7 Sonnet | 최대 128k 토큰 | 복잡한 추론, 코딩 및 AI 에이전트에 적합한 고성능 모델 |
| Claude 3.5 Sonnet v2 | 최대 200k 토큰 | 소프트웨어 공학, 에이전트 기능, 컴퓨터 상호작용에 특화된 최신 모델 |
| Claude 3.5 Sonnet | 최대 200k 토큰 | 다양한 작업에 탁월한 지능 및 추론 제공, 최적의 속도·비용 모델 |
| Claude 3.5 Haiku | 최대 200k 토큰 | 빠르고 컴팩트한 멀티모달 모델, 신속하고 자연스러운 대화에 최적 |
| Claude 3 Sonnet | 최대 200k 토큰 | 지능과 속도의 균형 잡힌 멀티모달 모델, 대규모 배포에 적합 |
| Claude 3 Haiku | 최대 200k 토큰 | 컴팩트한 고속 멀티모달 모델, 신속한 응답과 자연스러운 대화형 상호작용 |
| Claude 3 Opus | 최대 200k 토큰 | 인간 같은 추론과 우수한 문맥 이해로 복잡한 작업 수행 |
| Claude 2.1 | 최대 200k 토큰 | 확장된 컨텍스트, 신뢰도 개선, 로봇화 감소, 장문 및 RAG 적용에 적합 |
| Claude | 최대 100k 토큰 | 복잡한 대화, 창의적 콘텐츠 생성, 정교한 지시 수행에 탁월 |
| Claude Instant | 최대 100k 토큰 | 일상 대화, 분석, 요약, 문서 Q&A 등 빠르고 비용 효율적인 모델 |
| Llama 3.1 405B Instruct | 최대 128k 토큰 | 챗봇, 코딩, 도메인 특화 작업을 위한 합성 데이터 생성 및 추론용 첨단 LLM |
| Llama 3.1 70B Instruct | 최대 128k 토큰 | 복잡한 대화, 우수한 문맥 및 추론, 텍스트 생성 능력 강화 |
| Llama 3.1 8B Instruct | 최대 128k 토큰 | 우수한 언어 이해, 추론, 텍스트 생성 기능의 최첨단 모델 |
| Llama 3 70B Instruct | 최대 8k 토큰 | 복잡한 대화, 우수한 문맥 및 추론, 텍스트 생성 기능 강화 |
| Llama 3 8B Instruct | 최대 8k 토큰 | 첨단 언어 이해력, 추론, 텍스트 생성이 가능한 최첨단 LLM |
| Titan Text G1 - Lite | 최대 4k 토큰 | 영어 과제 및 요약, 콘텐츠 생성에 최적화된 경량 비용 효율적 모델 |
| Titan Text G1 - Express | 최대 8k 토큰 | 일반 언어, 대화, RAG 지원, 영어 및 100여 개 언어 지원 |
| Cohere Command | 최대 4k 토큰 | 사용자의 명령 수행, 실질적 기업 솔루션 제공에 특화된 모델 |
| Jurassic-2 Mid | 최대 8,191 토큰 | 다양한 언어 과제(Q&A, 요약, 생성 등)에 적합한 품질-비용 균형 모델 |
| Jurassic-2 Ultra | 최대 8,191 토큰 | 고급 텍스트 생성과 이해, 분석 및 콘텐츠 제작 등 복잡한 작업 수행 |
| Jamba-Instruct | 최대 256k 토큰 | 비용 효율적인 대용량 문맥 창작, 요약, Q&A에 최적화된 모델 |
| Mistral 7B Instruct | 최대 32k 토큰 | 명령을 따르고, 요청을 완성하며, 창의적 텍스트를 생성하는 LLM |
| Mistral 8x7B Instruct | 최대 32k 토큰 | 명령 및 요청 완성, 창의적 텍스트 생성이 가능한 MOE LLM |
</Accordion>
@@ -543,81 +445,13 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
CrewAI 프로젝트에서의 예시 사용법:
```python Code
llm = LLM(
model="nvidia_nim/meta/llama3-70b-instruct",
model="nvidia_nim/nvidia/nvidia-nemotron-3-ultra-550b-a55b",
temperature=0.7
)
```
Nvidia NIM은 일반 목적 작업부터 특수 목적 응용까지 다양한 용도를 위한 모델 제품군을 제공합니다.
NVIDIA NIM의 호스팅 카탈로그는 자주 변경됩니다. 현재 endpoint를 선택하고 모델 ID, 모달리티, 컨텍스트 제한을 확인하려면 [NVIDIA NIM 모델 카탈로그](https://build.nvidia.com/models)를 사용하세요.
| 모델 | 컨텍스트 윈도우 | 최적 용도 |
|------------------------------------------------------------------------|----------------|---------------------------------------------------------------------|
| nvidia/mistral-nemo-minitron-8b-8k-instruct | 8,192 토큰 | 챗봇, 가상 비서, 콘텐츠 생성을 위한 최신형 소형 언어 모델 |
| nvidia/nemotron-4-mini-hindi-4b-instruct | 4,096 토큰 | 힌디-영어 SLM, 힌디 언어 전용 온디바이스 추론 |
| nvidia/llama-3.1-nemotron-70b-instruct | 128k 토큰 | 더욱 도움이 되는 답변을 위해 커스터마이즈됨 |
| nvidia/llama3-chatqa-1.5-8b | 128k 토큰 | 챗봇, 검색엔진용 맥락 인식 응답 생성에 탁월한 고급 LLM |
| nvidia/llama3-chatqa-1.5-70b | 128k 토큰 | 챗봇, 검색엔진용 맥락 인식 응답 생성에 탁월한 고급 LLM |
| nvidia/vila | 128k 토큰 | 텍스트/이미지/비디오 이해 및 정보성 응답 생성을 지원하는 멀티모달 모델 |
| nvidia/neva-22 | 4,096 토큰 | 텍스트/이미지 이해 및 정보성 응답 생성을 지원하는 멀티모달 모델 |
| nvidia/nemotron-mini-4b-instruct | 8,192 토큰 | 일반 목적 작업 |
| nvidia/usdcode-llama3-70b-instruct | 128k 토큰 | OpenUSD 지식 질의 응답, USD-Python 코드 생성이 가능한 최신 LLM |
| nvidia/nemotron-4-340b-instruct | 4,096 토큰 | 실제 데이터를 모사하는 다양한 합성 데이터 생성 |
| meta/codellama-70b | 100k 토큰 | 자연어 → 코드 및 코드 → 자연어 전환 가능한 LLM |
| meta/llama2-70b | 4,096 토큰 | 텍스트, 코드 생성에 최적화된 최첨단 대형 언어 모델 |
| meta/llama3-8b-instruct | 8,192 토큰 | 최첨단 언어 이해 및 추론, 텍스트 생성 기능 모델 |
| meta/llama3-70b-instruct | 8,192 토큰 | 복잡한 대화, 우수한 문맥 및 추론, 텍스트 생성 |
| meta/llama-3.1-8b-instruct | 128k 토큰 | 최첨단 언어 이해 및 추론, 텍스트 생성 기능의 첨단 모델 |
| meta/llama-3.1-70b-instruct | 128k 토큰 | 복잡한 대화, 우수한 문맥 및 추론, 텍스트 생성 |
| meta/llama-3.1-405b-instruct | 128k 토큰 | 챗봇, 코딩, 도메인 특화 작업 합성 데이터 생성 및 추론 |
| meta/llama-3.2-1b-instruct | 128k 토큰 | 최첨단 소형 언어 이해, 추론, 텍스트 생성 모델 |
| meta/llama-3.2-3b-instruct | 128k 토큰 | 최첨단 소형 언어 이해, 추론, 텍스트 생성 |
| meta/llama-3.2-11b-vision-instruct | 128k 토큰 | 최첨단 소형 언어 이해, 추론, 텍스트 생성 |
| meta/llama-3.2-90b-vision-instruct | 128k 토큰 | 최첨단 소형 언어 이해, 추론, 텍스트 생성 |
| google/gemma-7b | 8,192 토큰 | 문자열의 이해, 변환, 코드 생성을 지원하는 최첨단 텍스트 생성 모델 |
| google/gemma-2b | 8,192 토큰 | 문자열의 이해, 변환, 코드 생성을 지원하는 최첨단 텍스트 생성 모델 |
| google/codegemma-7b | 8,192 토큰 | 코드 생성 및 보완에 특화된 Google Gemma-7B 기반 모델 |
| google/codegemma-1.1-7b | 8,192 토큰 | 코드 생성, 보완, 추론, 명령 수행에 강점을 가진 고급 프로그래밍 모델 |
| google/recurrentgemma-2b | 8,192 토큰 | 긴 시퀀스 생성 시 빠른 추론을 가능케 하는 순환 아키텍처 LLM |
| google/gemma-2-9b-it | 8,192 토큰 | 문자열의 이해, 변환, 코드 생성을 지원하는 최첨단 텍스트 생성 모델 |
| google/gemma-2-27b-it | 8,192 토큰 | 문자열의 이해, 변환, 코드 생성을 지원하는 최첨단 텍스트 생성 모델 |
| google/gemma-2-2b-it | 8,192 토큰 | 문자열의 이해, 변환, 코드 생성을 지원하는 최첨단 텍스트 생성 모델 |
| google/deplot | 512 토큰 | 플롯 이미지를 표로 변환하는 원샷 비주얼 언어 이해 모델 |
| google/paligemma | 8,192 토큰 | 텍스트, 이미지 입력 이해 및 정보성 응답 생성에 능한 비전 언어 모델 |
| mistralai/mistral-7b-instruct-v0.2 | 32k 토큰 | 명령을 따르고, 요청을 완성하며, 창의적 텍스트 생성이 가능한 LLM |
| mistralai/mixtral-8x7b-instruct-v0.1 | 8,192 토큰 | 명령 및 요청 완성, 창의 텍스트 생성이 가능한 MOE LLM |
| mistralai/mistral-large | 4,096 토큰 | 실제 데이터 특성을 모방하는 다양한 합성 데이터 생성 |
| mistralai/mixtral-8x22b-instruct-v0.1 | 8,192 토큰 | 실제 데이터 특성을 모방하는 다양한 합성 데이터 생성 |
| mistralai/mistral-7b-instruct-v0.3 | 32k 토큰 | 명령을 따르고, 요청을 완성하며, 창의적 텍스트 생성이 가능한 LLM |
| nv-mistralai/mistral-nemo-12b-instruct | 128k 토큰 | 추론, 코드, 다국어 작업에 적합한 최첨단 언어 모델; 단일 GPU에서 구동 |
| mistralai/mamba-codestral-7b-v0.1 | 256k 토큰 | 광범위한 프로그래밍 언어 및 작업에서 코드 작성 및 상호작용 전용 모델 |
| microsoft/phi-3-mini-128k-instruct | 128K 토큰 | 수학·논리 추론에 강한 경량 최신 공개 LLM |
| microsoft/phi-3-mini-4k-instruct | 4,096 토큰 | 수학·논리 추론에 강한 경량 최신 공개 LLM |
| microsoft/phi-3-small-8k-instruct | 8,192 토큰 | 수학·논리 추론에 강한 경량 최신 공개 LLM |
| microsoft/phi-3-small-128k-instruct | 128K 토큰 | 수학·논리 추론에 강한 경량 최신 공개 LLM |
| microsoft/phi-3-medium-4k-instruct | 4,096 토큰 | 수학·논리 추론에 강한 경량 최신 공개 LLM |
| microsoft/phi-3-medium-128k-instruct | 128K 토큰 | 수학·논리 추론에 강한 경량 최신 공개 LLM |
| microsoft/phi-3.5-mini-instruct | 128K 토큰 | 지연, 메모리/컴퓨트 한계 환경에서 AI 응용프로그램 구동 가능한 다국어 LLM |
| microsoft/phi-3.5-moe-instruct | 128K 토큰 | 연산 효율적 콘텐츠 생성을 위한 Mixture of Experts 기반 첨단 LLM |
| microsoft/kosmos-2 | 1,024 토큰 | 이미지의 시각적 요소 이해 및 추론을 위한 획기적 멀티모달 모델 |
| microsoft/phi-3-vision-128k-instruct | 128k 토큰 | 이미지에서 고품질 추론이 가능한 최첨단 공개 멀티모달 모델 |
| microsoft/phi-3.5-vision-instruct | 128k 토큰 | 이미지에서 고품질 추론이 가능한 최첨단 공개 멀티모달 모델 |
| databricks/dbrx-instruct | 12k 토큰 | 언어 이해, 코딩, RAG에 최신 성능을 제공하는 범용 LLM |
| snowflake/arctic | 1,024 토큰 | SQL 생성 및 코딩에 집중한 기업용 고효율 추론 모델 |
| aisingapore/sea-lion-7b-instruct | 4,096 토큰 | 동남아 언어 및 문화 다양성을 반영하는 LLM |
| ibm/granite-8b-code-instruct | 4,096 토큰 | 소프트웨어 프로그래밍 LLM, 코드 생성, 완성, 설명, 멀티턴 전환 |
| ibm/granite-34b-code-instruct | 8,192 토큰 | 소프트웨어 프로그래밍 LLM, 코드 생성, 완성, 설명, 멀티턴 전환 |
| ibm/granite-3.0-8b-instruct | 4,096 토큰 | RAG, 요약, 분류, 코드, 에이전틱AI 지원 첨단 소형 언어 모델 |
| ibm/granite-3.0-3b-a800m-instruct | 4,096 토큰 | RAG, 요약, 엔터티 추출, 분류에 최적화된 고효율 Mixture of Experts 모델 |
| mediatek/breeze-7b-instruct | 4,096 토큰 | 실제 데이터 특성을 모방하는 다양한 합성 데이터 생성 |
| upstage/solar-10.7b-instruct | 4,096 토큰 | 지시 따르기, 추론, 수학 등에서 뛰어난 NLP 작업 수행 |
| writer/palmyra-med-70b-32k | 32k 토큰 | 의료 분야에서 정확하고 문맥에 맞는 응답 생성에 선도적인 LLM |
| writer/palmyra-med-70b | 32k 토큰 | 의료 분야에서 정확하고 문맥에 맞는 응답 생성에 선도적인 LLM |
| writer/palmyra-fin-70b-32k | 32k 토큰 | 금융 분석, 보고, 데이터 처리에 특화된 LLM |
| 01-ai/yi-large | 32k 토큰 | 영어, 중국어로 훈련, 챗봇 및 창의적 글쓰기 등 다양한 작업에 사용 |
| deepseek-ai/deepseek-coder-6.7b-instruct | 2k 토큰 | 고급 코드 생성, 완성, 인필링 등의 기능을 제공하는 강력한 코딩 모델 |
| rakuten/rakutenai-7b-instruct | 1,024 토큰 | 언어 이해, 추론, 텍스트 생성이 탁월한 최첨단 LLM |
| rakuten/rakutenai-7b-chat | 1,024 토큰 | 언어 이해, 추론, 텍스트 생성이 탁월한 최첨단 LLM |
| baichuan-inc/baichuan2-13b-chat | 4,096 토큰 | 중국어 및 영어 대화, 코딩, 수학, 지시 따르기, 퀴즈 풀이 지원 |
**참고:** 이 제공자는 LiteLLM을 사용합니다. 프로젝트에 의존성으로 추가하세요:
```bash
@@ -680,15 +514,12 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
CrewAI 프로젝트에서의 예시 사용법:
```python Code
llm = LLM(
model="groq/llama-3.2-90b-text-preview",
model="groq/qwen/qwen3.6-27b",
temperature=0.7
)
```
| 모델 | 컨텍스트 윈도우 | 최적 용도 |
|-----------------|-------------------|----------------------------------|
| Llama 3.1 70B/8B| 131,072 토큰 | 고성능, 대용량 문맥 작업 |
| Llama 3.2 Series| 8,192 토큰 | 범용 작업 |
| Mixtral 8x7B | 32,768 토큰 | 성능과 문맥의 균형 |
Groq는 production 모델과 preview 모델을 구분하며 모델 ID를 정기적으로 폐기합니다. 프로덕션 모델을 선택하기 전에 [Groq 모델 카탈로그](https://console.groq.com/docs/models)와 [지원 중단 페이지](https://console.groq.com/docs/deprecations)를 확인하세요.
**참고:** 이 제공자는 LiteLLM을 사용합니다. 프로젝트에 의존성으로 추가하세요:
```bash
@@ -770,11 +601,12 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
CrewAI 프로젝트에서의 예시 사용법:
```python Code
llm = LLM(
model="llama-3.1-sonar-large-128k-online",
base_url="https://api.perplexity.ai/"
model="perplexity/sonar-pro"
)
```
현재 모델 ID와 지원 중단 공지는 [Perplexity 모델 카탈로그](https://docs.perplexity.ai/getting-started/models)와 [changelog](https://docs.perplexity.ai/docs/resources/changelog)를 확인하세요.
**참고:** 이 제공자는 LiteLLM을 사용합니다. 프로젝트에 의존성으로 추가하세요:
```bash
uv add 'crewai[litellm]'
@@ -810,17 +642,12 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
CrewAI 프로젝트에서의 예시 사용법:
```python Code
llm = LLM(
model="sambanova/Meta-Llama-3.1-8B-Instruct",
model="sambanova/Meta-Llama-3.3-70B-Instruct",
temperature=0.7
)
```
| 모델 | 컨텍스트 윈도우 | 최적 용도 |
|-----------------|---------------------|--------------------------------------|
| Llama 3.1 70B/8B| 최대 131,072 토큰 | 고성능, 대용량 문맥 작업 |
| Llama 3.1 405B | 8,192 토큰 | 고성능, 높은 출력 품질 |
| Llama 3.2 Series| 8,192 토큰 | 범용, 멀티모달 작업 |
| Llama 3.3 70B | 최대 131,072 토큰 | 고성능, 높은 출력 품질 |
| Qwen2 familly | 8,192 토큰 | 고성능, 높은 출력 품질 |
SambaNova Cloud의 호스팅 모델은 CrewAI와 별도로 변경될 수 있습니다. 배포 전에 [models endpoint](https://docs.sambanova.ai/docs/api-reference/models/get-environments-available-model-list-metadata)를 조회하고 [지원 중단 가이드](https://docs.sambanova.ai/docs/en/models/deprecations)를 확인하세요.
**참고:** 이 제공자는 LiteLLM을 사용합니다. 프로젝트에 의존성으로 추가하세요:
```bash
@@ -838,7 +665,7 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
CrewAI 프로젝트에서의 예시 사용법:
```python Code
llm = LLM(
model="cerebras/llama3.1-70b",
model="cerebras/gpt-oss-120b",
temperature=0.7,
max_tokens=8192
)
@@ -852,6 +679,8 @@ CrewAI는 고유한 기능, 인증 방법, 모델 역량을 제공하는 다양
- 긴 컨텍스트 윈도우 지원
</Info>
현재 공개 endpoint ID는 [Cerebras 모델 카탈로그](https://inference-docs.cerebras.ai/models/overview)와 [지원 중단 공지](https://inference-docs.cerebras.ai/support/deprecation)를 확인하세요.
**참고:** 이 제공자는 LiteLLM을 사용합니다. 프로젝트에 의존성으로 추가하세요:
```bash
uv add 'crewai[litellm]'
@@ -926,7 +755,7 @@ CrewAI는 LLM의 스트리밍 응답을 지원하여, 애플리케이션이 출
# 스트리밍이 활성화된 LLM 생성
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
stream=True # 스트리밍 활성화
)
```
@@ -976,7 +805,7 @@ CrewAI는 LLM의 스트리밍 응답을 지원하여, 애플리케이션이 출
my_listener = MyCustomListener()
llm = LLM(model="gpt-4o-mini", temperature=0, stream=True)
llm = LLM(model="openai/gpt-5.6-terra", stream=True)
researcher = Agent(
role="About User",
@@ -1012,6 +841,8 @@ CrewAI는 LLM의 스트리밍 응답을 지원하여, 애플리케이션이 출
CrewAI는 Pydantic 모델을 사용하여 `response_format`을 정의함으로써 LLM 호출에서 구조화된 응답을 지원합니다. 이를 통해 프레임워크가 출력을 자동으로 파싱하고 검증할 수 있어, 수동 후처리 없이도 응답을 애플리케이션에 쉽게 통합할 수 있습니다.
구조화된 출력 지원은 제공업체와 모델에 따라 다릅니다. 프로덕션에서 구조화된 응답에 의존하기 전에 선택한 모델을 테스트하세요.
예를 들어, 예상되는 응답 구조를 나타내는 Pydantic 모델을 정의하고 LLM을 인스턴스화할 때 `response_format`으로 전달할 수 있습니다. 이 모델은 LLM 출력을 구조화된 Python 객체로 변환하는 데 사용됩니다.
```python Code
@@ -1023,7 +854,7 @@ class Dog(BaseModel):
breed: str
llm = LLM(model="gpt-4o", response_format=Dog)
llm = LLM(model="openai/gpt-5.6-terra", response_format=Dog)
response = llm.call(
"Analyze the following messages and return the name, age, and breed. "
@@ -1052,8 +883,8 @@ LLM 설정을 최대한 활용하는 방법을 알아보세요:
# 3. 큰 컨텍스트에 대한 작업 분할
llm = LLM(
model="gpt-4",
max_tokens=4000, # 응답 길이 제한
model="openai/gpt-5.6-terra",
max_completion_tokens=4000, # 응답 길이 제한
)
```
@@ -1077,15 +908,14 @@ LLM 설정을 최대한 활용하는 방법을 알아보세요:
```python
# 모델을 적절한 설정으로 구성
llm = LLM(
model="openai/gpt-4-turbo-preview",
temperature=0.7, # 작업에 따라 조정
max_tokens=4096, # 출력 요구 사항에 맞게 설정
timeout=300 # 복잡한 작업을 위한 더 긴 타임아웃
model="openai/gpt-5.6-terra",
reasoning_effort="medium",
max_completion_tokens=4096,
timeout=300
)
```
<Tip>
- 사실 기반 응답에는 낮은 temperature(0.1~0.3)
- 창의적인 작업에는 높은 temperature(0.7~0.9)
선택한 모델이 지원하는 제어 옵션을 사용하세요. 제공자에 따라 `temperature`, reasoning 또는 thinking 수준, 혹은 원하는 스타일과 변동성을 정의하는 프롬프트 지침을 사용할 수 있습니다.
</Tip>
</Step>

292
docs/edge/pt-BR/concepts/llms.mdx
View File

@@ -21,7 +21,7 @@ Modelos de Linguagem de Grande Escala (LLMs) são a inteligência central por tr
A janela de contexto determina quanto texto um LLM pode processar de uma só vez. Janelas maiores (por exemplo, 128K tokens) permitem mais contexto, porém podem ser mais caras e lentas.
</Card>
<Card title="Temperatura" icon="temperature-three-quarters">
A temperatura (0.0 a 1.0) controla a aleatoriedade das respostas. Valores mais baixos (ex.: 0.2) produzem respostas mais focadas e determinísticas, enquanto valores mais altos (ex.: 0.8) aumentam criatividade e variabilidade.
A temperatura é um controle de amostragem compatível com alguns modelos. Valores mais baixos geralmente tornam a amostragem mais focada, enquanto valores mais altos aumentam a variabilidade. Alguns modelos de raciocínio mais recentes ignoram, desaconselham ou rejeitam esse parâmetro; consulte a documentação do modelo escolhido antes de defini-lo.
</Card>
<Card title="Seleção de Provedor" icon="server">
Cada provedor de LLM (ex.: OpenAI, Anthropic, Google) oferece modelos diferentes, com capacidades, preços e recursos variados. Escolha conforme suas necessidades de precisão, velocidade e custo.
@@ -37,7 +37,7 @@ Existem diferentes locais no código do CrewAI onde você pode especificar o mod
A maneira mais simples de começar. Defina o modelo diretamente em seu ambiente, usando um arquivo `.env` ou no código do seu aplicativo. Se você utilizou `crewai create` para iniciar seu projeto, já estará configurado.
```bash .env
MODEL=model-id # e.g. gpt-4o, gemini-2.0-flash, claude-3-sonnet-...
MODEL=provider/model-id # e.g. openai/gpt-5.6-terra
# Lembre-se de definir suas chaves de API aqui também. Veja a seção
# do Provedor abaixo.
@@ -56,7 +56,7 @@ Existem diferentes locais no código do CrewAI onde você pode especificar o mod
goal: Conduct comprehensive research and analysis
backstory: A dedicated research professional with years of experience
verbose: true
llm: provider/model-id # e.g. openai/gpt-4o, google/gemini-2.0-flash, anthropic/claude...
llm: provider/model-id # e.g. anthropic/claude-sonnet-4-6
# (veja exemplos de configuração de provedores abaixo para mais)
```
@@ -75,32 +75,24 @@ Existem diferentes locais no código do CrewAI onde você pode especificar o mod
from crewai import LLM
# Configuração básica
llm = LLM(model="model-id-here") # gpt-4o, gemini-2.0-flash, anthropic/claude...
llm = LLM(model="provider/model-id") # e.g. gemini/gemini-3.6-flash
# Configuração avançada com parâmetros detalhados
llm = LLM(
model="openai/gpt-4",
temperature=0.8,
max_tokens=150,
top_p=0.9,
frequency_penalty=0.1,
presence_penalty=0.1,
response_format={"type":"json"},
stop=["FIM"],
seed=42
model="provider/model-id",
timeout=120,
max_tokens=4000,
response_format={"type": "json"}, # Para saídas estruturadas
)
```
<Info>
Explicações dos parâmetros:
- `temperature`: Controla a aleatoriedade (0.0-1.0)
- `timeout`: Tempo máximo de espera pela resposta
- `max_tokens`: Limita o comprimento da resposta
- `top_p`: Alternativa à temperatura para amostragem
- `frequency_penalty`: Reduz repetição de palavras
- `presence_penalty`: Incentiva novos tópicos
- `response_format`: Especifica formato de saída
- `seed`: Garante resultados consistentes
Controles de amostragem como `temperature` e `top_p`, parâmetros de penalidade, nomes de limites de tokens e controles de raciocínio são específicos de cada modelo. Adicione-os somente quando o provedor e o modelo escolhidos oferecerem suporte. Consulte os exemplos de provedores abaixo e a documentação do modelo do provedor.
</Info>
</Tab>
</Tabs>
@@ -119,6 +111,10 @@ Existem diferentes locais no código do CrewAI onde você pode especificar o mod
O CrewAI suporta uma grande variedade de provedores de LLM, cada um com recursos, métodos de autenticação e capacidades de modelo únicos.
Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, configurar e otimizar o LLM que melhor atende às necessidades do seu projeto.
<Warning>
A disponibilidade dos modelos muda com frequência e pode variar por conta, região e plataforma de nuvem. Os exemplos abaixo usam modelos atuais no momento da redação, mas não são listas completas de suporte. Antes de implantar, confirme o ID e o estado do ciclo de vida do modelo no catálogo vinculado do provedor.
</Warning>
<AccordionGroup>
<Accordion title="OpenAI">
Defina as seguintes variáveis de ambiente no seu arquivo `.env`:
@@ -137,28 +133,13 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
from crewai import LLM
llm = LLM(
model="openai/gpt-4",
temperature=0.8,
max_tokens=150,
top_p=0.9,
frequency_penalty=0.1,
presence_penalty=0.1,
stop=["FIM"],
seed=42
model="openai/gpt-5.6-terra",
reasoning_effort="medium",
max_completion_tokens=4000
)
```
OpenAI é um dos líderes em modelos LLM com uma ampla gama de modelos e recursos.
| Modelo | Janela de Contexto | Melhor Para |
|----------------------|---------------------|------------------------------------------|
| GPT-4 | 8.192 tokens | Tarefas de alta precisão, raciocínio complexo |
| GPT-4 Turbo | 128.000 tokens | Conteúdo longo, análise de documentos |
| GPT-4o & GPT-4o-mini | 128.000 tokens | Processamento de contexto amplo com bom custo-benefício |
| o3-mini | 200.000 tokens | Raciocínio rápido, tarefas complexas |
| o1-mini | 128.000 tokens | Raciocínio rápido, tarefas complexas |
| o1-preview | 128.000 tokens | Raciocínio rápido, tarefas complexas |
| o1 | 200.000 tokens | Raciocínio rápido, tarefas complexas |
A OpenAI adiciona modelos e desativa snapshots antigos regularmente. Consulte o [catálogo de modelos da OpenAI](https://developers.openai.com/api/docs/models) para obter IDs atuais, janelas de contexto, compatibilidade com endpoints e informações de ciclo de vida.
**Responses API:**
@@ -215,14 +196,7 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
)
```
Todos os modelos listados em https://llama.developer.meta.com/docs/models/ são suportados.
| ID do Modelo | Comprimento contexto entrada | Comprimento contexto saída | Modalidades de entrada | Modalidades de saída |
| --- | --- | --- | --- | --- |
| `meta_llama/Llama-4-Scout-17B-16E-Instruct-FP8` | 128k | 4028 | Texto, Imagem | Texto |
| `meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8` | 128k | 4028 | Texto, Imagem | Texto |
| `meta_llama/Llama-3.3-70B-Instruct` | 128k | 4028 | Texto | Texto |
| `meta_llama/Llama-3.3-8B-Instruct` | 128k | 4028 | Texto | Texto |
Consulte a [visão geral dos modelos Meta Llama](https://ai.meta.com/llama/get-started/) para conhecer as famílias de modelos, modalidades e orientações de contexto atuais.
**Nota:** Este provedor usa o LiteLLM. Adicione-o como dependência ao seu projeto:
```bash
@@ -291,10 +265,12 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
Exemplo de uso em seu projeto CrewAI:
```python Code
llm = LLM(
model="anthropic/claude-3-sonnet-20240229-v1:0",
temperature=0.7
model="anthropic/claude-sonnet-4-6",
max_tokens=4096
)
```
Consulte a [visão geral dos modelos](https://platform.claude.com/docs/en/about-claude/models/overview) da Anthropic para obter IDs e capacidades atuais e revise a [tabela de descontinuação](https://platform.claude.com/docs/en/about-claude/model-deprecations) antes de fixar um modelo em produção.
</Accordion>
<Accordion title="Google (Gemini API)">
@@ -319,8 +295,7 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.0-flash",
temperature=0.7,
model="gemini/gemini-3.6-flash",
)
```
@@ -339,8 +314,7 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
from crewai import LLM
llm = LLM(
model="gemini/gemini-2.0-flash",
temperature=0.7
model="gemini/gemini-3.6-flash"
)
```
@@ -352,47 +326,15 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
Para mais detalhes, consulte a [documentação do Vertex AI Express mode](https://docs.cloud.google.com/vertex-ai/generative-ai/docs/start/quickstart?usertype=apikey).
</Info>
### Modelos Gemini
O Google oferece uma variedade de modelos poderosos otimizados para diferentes casos de uso.
| Modelo | Janela de Contexto | Melhor Para |
|----------------------------------|--------------------|---------------------------------------------------------------------|
| gemini-2.5-flash-preview-04-17 | 1M tokens | Pensamento adaptativo, eficiência de custo |
| gemini-2.5-pro-preview-05-06 | 1M tokens | Pensamento e raciocínio avançados, compreensão multimodal, codificação avançada, etc. |
| gemini-2.0-flash | 1M tokens | Próxima geração de recursos, velocidade, raciocínio e streaming em tempo real |
| gemini-2.0-flash-lite | 1M tokens | Eficiência de custo e baixa latência |
| gemini-1.5-flash | 1M tokens | Modelo multimodal equilibrado, bom para maioria das tarefas |
| gemini-1.5-flash-8B | 1M tokens | Mais rápido, mais eficiente em custo, adequado para tarefas de alta frequência |
| gemini-1.5-pro | 2M tokens | Melhor desempenho para uma ampla variedade de tarefas de raciocínio, incluindo lógica, codificação e colaboração criativa |
A lista completa de modelos está disponível na [documentação dos modelos Gemini](https://ai.google.dev/gemini-api/docs/models).
### Gemma
A API Gemini também permite uso de sua chave de API para acessar [modelos Gemma](https://ai.google.dev/gemma/docs) hospedados na infraestrutura Google.
| Modelo | Janela de Contexto |
|----------------|-------------------|
| gemma-3-1b-it | 32k tokens |
| gemma-3-4b-it | 32k tokens |
| gemma-3-12b-it | 32k tokens |
| gemma-3-27b-it | 128k tokens |
O Google publica IDs atuais, capacidades e estágios do ciclo de vida no [catálogo de modelos Gemini](https://ai.google.dev/gemini-api/docs/models). Consulte o [cronograma de descontinuação](https://ai.google.dev/gemini-api/docs/deprecations) antes de escolher um modelo estável ou preview. A API Gemini também hospeda [modelos Gemma](https://ai.google.dev/gemma/docs).
</Accordion>
<Accordion title="Google (Vertex AI)">
Obtenha as credenciais pelo Google Cloud Console, salve em um arquivo JSON e carregue com o código a seguir:
```python Code
import json
file_path = 'path/to/vertex_ai_service_account.json'
# Carregar o arquivo JSON
with open(file_path, 'r') as file:
vertex_credentials = json.load(file)
# Converter credenciais em string JSON
vertex_credentials_json = json.dumps(vertex_credentials)
Autentique-se com as [Credenciais Padrão do Aplicativo](https://cloud.google.com/docs/authentication/provide-credentials-adc) e configure o provedor Gemini nativo para usar o Vertex AI:
```toml .env
GOOGLE_GENAI_USE_VERTEXAI=true
GOOGLE_CLOUD_PROJECT=<your-project-id>
GOOGLE_CLOUD_LOCATION=<location>
```
Exemplo de uso em seu projeto CrewAI:
@@ -400,27 +342,15 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
from crewai import LLM
llm = LLM(
model="gemini-1.5-pro-latest", # or vertex_ai/gemini-1.5-pro-latest
temperature=0.7,
vertex_credentials=vertex_credentials_json
model="gemini/gemini-3.6-flash"
)
```
O Google oferece uma variedade de modelos poderosos otimizados para diferentes casos de uso:
Consulte as [informações de modelos do Vertex AI](https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models) para verificar modelos e regiões disponíveis.
| Modelo | Janela de Contexto | Melhor Para |
|----------------------------------|--------------------|---------------------------------------------------------------------|
| gemini-2.5-flash-preview-04-17 | 1M tokens | Pensamento adaptativo, eficiência de custo |
| gemini-2.5-pro-preview-05-06 | 1M tokens | Pensamento e raciocínio avançados, compreensão multimodal, codificação avançada, etc. |
| gemini-2.0-flash | 1M tokens | Próxima geração de recursos, velocidade, raciocínio e streaming em tempo real |
| gemini-2.0-flash-lite | 1M tokens | Eficiência de custo e baixa latência |
| gemini-1.5-flash | 1M tokens | Modelo multimodal equilibrado, bom para maioria das tarefas |
| gemini-1.5-flash-8B | 1M tokens | Mais rápido, mais eficiente em custo, adequado para tarefas de alta frequência |
| gemini-1.5-pro | 2M tokens | Melhor desempenho para uma ampla variedade de tarefas de raciocínio, incluindo lógica, codificação e colaboração criativa |
**Nota:** Este provedor usa o LiteLLM. Adicione-o como dependência ao seu projeto:
**Nota:** Esta configuração usa a integração Gemini nativa do CrewAI. Adicione-a como dependência ao seu projeto:
```bash
uv add 'crewai[litellm]'
uv add "crewai[google-genai]"
```
</Accordion>
@@ -455,7 +385,7 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
Exemplo de uso em seu projeto CrewAI:
```python Code
llm = LLM(
model="bedrock/anthropic.claude-3-sonnet-20240229-v1:0"
model="bedrock/us.anthropic.claude-sonnet-4-6"
)
```
@@ -463,34 +393,6 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
[Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html) é um serviço gerenciado que fornece acesso a múltiplos modelos fundamentais dos principais provedores de IA através de uma API unificada, permitindo o desenvolvimento seguro e responsável de aplicações de IA.
| Modelo | Janela de Contexto | Melhor Para |
|--------------------------|------------------------|---------------------------------------------------------------------|
| Amazon Nova Pro | Até 300k tokens | Alto desempenho, equilíbrio entre precisão, velocidade e custo em tarefas diversas. |
| Amazon Nova Micro | Até 128k tokens | Modelo texto-only de alta performance, custo-benefício, otimizado para baixa latência. |
| Amazon Nova Lite | Até 300k tokens | Alto desempenho, processamento multimodal acessível para texto, imagem, vídeo em tempo real. |
| Claude 3.7 Sonnet | Até 128k tokens | Alto desempenho para raciocínio complexo, programação & agentes de IA|
| Claude 3.5 Sonnet v2 | Até 200k tokens | Modelo avançado especializado em engenharia de software, capacidades agenticas e interação computacional com custo otimizado. |
| Claude 3.5 Sonnet | Até 200k tokens | Alto desempenho com inteligência e raciocínio excepcionais, equilíbrio entre velocidade-custo. |
| Claude 3.5 Haiku | Até 200k tokens | Modelo multimodal rápido e compacto, otimizado para respostas rápidas e interações humanas naturais |
| Claude 3 Sonnet | Até 200k tokens | Modelo multimodal equilibrando inteligência e velocidade para grandes volumes de uso. |
| Claude 3 Haiku | Até 200k tokens | Compacto, multimodal, otimizado para respostas rápidas e diálogo natural |
| Claude 3 Opus | Até 200k tokens | Modelo multimodal mais avançado para tarefas complexas com raciocínio humano e entendimento contextual superior. |
| Claude 2.1 | Até 200k tokens | Versão aprimorada com janela de contexto aumentada, maior confiabilidade, menos alucinações para aplicações longas e RAG |
| Claude | Até 100k tokens | Modelo versátil para diálogos sofisticados, conteúdo criativo e instruções precisas. |
| Claude Instant | Até 100k tokens | Modelo rápido e de baixo custo para tarefas diárias, como diálogos, análise, sumarização e Q&A em documentos |
| Llama 3.1 405B Instruct | Até 128k tokens | LLM avançado para geração de dados sintéticos, distilação e inferência para chatbots, programação, tarefas de domínio específico. |
| Llama 3.1 70B Instruct | Até 128k tokens | Potencializa conversas complexas com entendimento contextual superior, raciocínio e geração de texto. |
| Llama 3.1 8B Instruct | Até 128k tokens | Modelo de última geração, entendimento de linguagem, raciocínio e geração de texto. |
| Llama 3 70B Instruct | Até 8k tokens | Potencializa conversas complexas com entendimento contextual superior, raciocínio e geração de texto. |
| Llama 3 8B Instruct | Até 8k tokens | LLM de última geração com excelente desempenho em linguagem e geração de texto. |
| Titan Text G1 - Lite | Até 4k tokens | Modelo leve e econômico para tarefas em inglês e ajuste fino, focado em sumarização e geração de conteúdo. |
| Titan Text G1 - Express | Até 8k tokens | Modelo versátil para tarefas gerais de linguagem, chat e aplicações RAG com suporte a inglês e 100+ línguas. |
| Cohere Command | Até 4k tokens | Modelo especializado em seguir comandos do usuário e entregar soluções empresariais práticas. |
| Jurassic-2 Mid | Até 8.191 tokens | Modelo econômico equilibrando qualidade e custo para tarefas como Q&A, sumarização e geração de conteúdo. |
| Jurassic-2 Ultra | Até 8.191 tokens | Geração avançada de texto e compreensão, excelente em análise e criação de conteúdo complexo. |
| Jamba-Instruct | Até 256k tokens | Modelo com janela de contexto extendida para geração de texto, sumarização e Q&A de baixo custo. |
| Mistral 7B Instruct | Até 32k tokens | LLM atende instruções, solicitações e gera texto criativo. |
| Mistral 8x7B Instruct | Até 32k tokens | MOE LLM que atende instruções, solicitações e gera texto criativo. |
</Accordion>
<Accordion title="Amazon SageMaker">
@@ -542,81 +444,13 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
Exemplo de uso em seu projeto CrewAI:
```python Code
llm = LLM(
model="nvidia_nim/meta/llama3-70b-instruct",
model="nvidia_nim/nvidia/nvidia-nemotron-3-ultra-550b-a55b",
temperature=0.7
)
```
O Nvidia NIM oferece uma suíte abrangente de modelos para diversos usos, desde tarefas gerais até aplicações especializadas.
O catálogo hospedado do NVIDIA NIM muda com frequência. Use o [catálogo de modelos NVIDIA NIM](https://build.nvidia.com/models) para escolher um endpoint atual e confirmar o ID, as modalidades e os limites de contexto.
| Modelo | Janela de Contexto | Melhor Para |
|--------------------------------------------------------------------------|--------------------|---------------------------------------------------------------------|
| nvidia/mistral-nemo-minitron-8b-8k-instruct | 8.192 tokens | Modelo pequeno de linguagem topo de linha para chatbots, assistentes virtuais e geração de conteúdo. |
| nvidia/nemotron-4-mini-hindi-4b-instruct | 4.096 tokens | SLM bilíngue Hindi-Inglês para inferência no dispositivo, específico para língua hindi. |
| nvidia/llama-3.1-nemotron-70b-instruct | 128k tokens | Personalizado para respostas mais úteis |
| nvidia/llama3-chatqa-1.5-8b | 128k tokens | LLM avançado para respostas contextuais de alta qualidade em chatbots e mecanismos de busca. |
| nvidia/llama3-chatqa-1.5-70b | 128k tokens | LLM avançado para respostas contextuais de alta qualidade para chatbots e mecanismos de busca. |
| nvidia/vila | 128k tokens | Modelo multmodal visão-linguagem para compreensão de texto/img/vídeo com respostas informativas |
| nvidia/neva-22 | 4.096 tokens | Modelo de visão-linguagem multimodal para compreensão textos/imagens e respostas informativas |
| nvidia/nemotron-mini-4b-instruct | 8.192 tokens | Tarefas gerais |
| nvidia/usdcode-llama3-70b-instruct | 128k tokens | LLM de ponta para queries OpenUSD e geração de código USD-Python. |
| nvidia/nemotron-4-340b-instruct | 4.096 tokens | Gera dados sintéticos diversos simulando características reais. |
| meta/codellama-70b | 100k tokens | LLM capaz de gerar código a partir de linguagem natural e vice-versa.|
| meta/llama2-70b | 4.096 tokens | Modelo de IA avançado para geração de textos e códigos. |
| meta/llama3-8b-instruct | 8.192 tokens | LLM de última geração, entendimento de linguagem, raciocínio e geração de texto. |
| meta/llama3-70b-instruct | 8.192 tokens | Potencializa conversas complexas com entendimento contextual superior, raciocínio e geração de texto.|
| meta/llama-3.1-8b-instruct | 128k tokens | Modelo compacto de última geração, com compreensão, raciocínio e geração de texto superior. |
| meta/llama-3.1-70b-instruct | 128k tokens | Potencializa conversas complexas com entendimento contextual superior, raciocínio e geração de texto. |
| meta/llama-3.1-405b-instruct | 128k tokens | LLM avançado para geração sintética de dados, destilação e inferência para chatbots, código, tarefas de domínio específico. |
| meta/llama-3.2-1b-instruct | 128k tokens | Pequeno modelo de linguagem de última geração, entendimento, raciocínio e geração textual.|
| meta/llama-3.2-3b-instruct | 128k tokens | Pequeno modelo de linguagem de última geração, entendimento, raciocínio e geração textual.|
| meta/llama-3.2-11b-vision-instruct | 128k tokens | Pequeno modelo de linguagem de última geração, entendimento, raciocínio e geração textual multimodal.|
| meta/llama-3.2-90b-vision-instruct | 128k tokens | Pequeno modelo de linguagem de última geração, entendimento, raciocínio e geração textual multimodal.|
| google/gemma-7b | 8.192 tokens | Modelo avançado de geração de texto, compreensão, transformação e programação.|
| google/gemma-2b | 8.192 tokens | Modelo avançado de geração de texto, compreensão, transformação e programação.|
| google/codegemma-7b | 8.192 tokens | Modelo avançado baseado no Gemma-7B do Google, especializado em geração de códigos e autocomplete.|
| google/codegemma-1.1-7b | 8.192 tokens | Modelo avançado para geração, complemento, raciocínio e instrução em código.|
| google/recurrentgemma-2b | 8.192 tokens | Modelo baseado em arquitetura recorrente para inferência mais rápida em sequências longas.|
| google/gemma-2-9b-it | 8.192 tokens | Modelo avançado de geração de texto, compreensão, transformação e programação.|
| google/gemma-2-27b-it | 8.192 tokens | Modelo avançado de geração de texto, compreensão, transformação e programação.|
| google/gemma-2-2b-it | 8.192 tokens | Modelo avançado de geração de texto, compreensão, transformação e programação.|
| google/deplot | 512 tokens | Modelo visual por linguagem para entender gráficos e converter em tabelas.|
| google/paligemma | 8.192 tokens | Modelo visão-linguagem experto em compreender texto e visual, gerando respostas informativas.|
| mistralai/mistral-7b-instruct-v0.2 | 32k tokens | LLM que segue instruções, completa pedidos e gera texto criativo. |
| mistralai/mixtral-8x7b-instruct-v0.1 | 8.192 tokens | MOE LLM para seguir instruções e gerar versões criativas de texto. |
| mistralai/mistral-large | 4.096 tokens | Geração de dados sintéticos. |
| mistralai/mixtral-8x22b-instruct-v0.1 | 8.192 tokens | Geração de dados sintéticos. |
| mistralai/mistral-7b-instruct-v0.3 | 32k tokens | LLM que segue instruções, completa pedidos e gera texto criativo. |
| nv-mistralai/mistral-nemo-12b-instruct | 128k tokens | Modelo de linguagem avançado para raciocínio, código, tarefas multilíngues; roda em uma única GPU.|
| mistralai/mamba-codestral-7b-v0.1 | 256k tokens | Modelo para escrita e interação com código em múltiplas linguagens e tarefas.|
| microsoft/phi-3-mini-128k-instruct | 128K tokens | LLM leve, de última geração, com habilidades de lógica e matemática.|
| microsoft/phi-3-mini-4k-instruct | 4.096 tokens | LLM leve, de última geração, com habilidades de lógica e matemática.|
| microsoft/phi-3-small-8k-instruct | 8.192 tokens | LLM leve, de última geração, com habilidades de lógica e matemática.|
| microsoft/phi-3-small-128k-instruct | 128K tokens | LLM leve, de última geração, com habilidades de lógica e matemática.|
| microsoft/phi-3-medium-4k-instruct | 4.096 tokens | LLM leve, de última geração, com habilidades de lógica e matemática.|
| microsoft/phi-3-medium-128k-instruct | 128K tokens | LLM leve, de última geração, com habilidades de lógica e matemática.|
| microsoft/phi-3.5-mini-instruct | 128K tokens | LLM multilíngue leve para aplicações de IA restritas em memória e tempo.|
| microsoft/phi-3.5-moe-instruct | 128K tokens | LLM avançada baseada em Mixture of Experts para geração eficiente de conteúdo.|
| microsoft/kosmos-2 | 1.024 tokens | Modelo multimodal revolucionário para compreender e raciocinar elementos visuais em imagens.|
| microsoft/phi-3-vision-128k-instruct | 128k tokens | Modelo multimodal aberto de ponta para raciocínio de alta qualidade a partir de imagens.|
| microsoft/phi-3.5-vision-instruct | 128k tokens | Modelo multimodal aberto de ponta para raciocínio de alta qualidade a partir de imagens.|
| databricks/dbrx-instruct | 12k tokens | LLM de uso geral com desempenho no estado da arte para linguagem, programação e RAG.|
| snowflake/arctic | 1.024 tokens | Inferência eficiente para aplicações empresariais focadas em SQL e programação.|
| aisingapore/sea-lion-7b-instruct | 4.096 tokens | LLM para representação e diversidade linguística e cultural do sudeste asiático.|
| ibm/granite-8b-code-instruct | 4.096 tokens | LLM para programação: geração, explicação e diálogo multi-turn de código.|
| ibm/granite-34b-code-instruct | 8.192 tokens | LLM para programação: geração, explicação e diálogo multi-turn de código.|
| ibm/granite-3.0-8b-instruct | 4.096 tokens | Pequeno modelo avançado, com suporte a RAG, sumário, classificação, código e IA agentica.|
| ibm/granite-3.0-3b-a800m-instruct | 4.096 tokens | Modelo Mixture of Experts eficiente para RAG, sumário, extração de entidades, classificação.|
| mediatek/breeze-7b-instruct | 4.096 tokens | Gera dados sintéticos diversos.|
| upstage/solar-10.7b-instruct | 4.096 tokens | Excelente em tarefas de PLN, especialmente seguir instruções, raciocínio e matemática.|
| writer/palmyra-med-70b-32k | 32k tokens | LLM líder para respostas médicas precisas e contextuais.|
| writer/palmyra-med-70b | 32k tokens | LLM líder para respostas médicas precisas e contextuais.|
| writer/palmyra-fin-70b-32k | 32k tokens | LLM especializada em análise financeira, relatórios e processamento de dados.|
| 01-ai/yi-large | 32k tokens | Poderoso para inglês e chinês, incluindo chatbot e escrita criativa.|
| deepseek-ai/deepseek-coder-6.7b-instruct | 2k tokens | Modelo avançado para geração de código, autocomplete, infilling.|
| rakuten/rakutenai-7b-instruct | 1.024 tokens | LLM topo de linha, compreensão, raciocínio e geração textual.|
| rakuten/rakutenai-7b-chat | 1.024 tokens | LLM topo de linha, compreensão, raciocínio e geração textual.|
| baichuan-inc/baichuan2-13b-chat | 4.096 tokens | Suporte a chat em chinês/inglês, programação, matemática, seguir instruções, resolver quizzes.|
**Nota:** Este provedor usa o LiteLLM. Adicione-o como dependência ao seu projeto:
```bash
@@ -679,15 +513,12 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
Exemplo de uso em seu projeto CrewAI:
```python Code
llm = LLM(
model="groq/llama-3.2-90b-text-preview",
model="groq/qwen/qwen3.6-27b",
temperature=0.7
)
```
| Modelo | Janela de Contexto | Melhor Para |
|-------------------|---------------------|------------------------------------------|
| Llama 3.1 70B/8B | 131.072 tokens | Alta performance e tarefas de contexto grande|
| Llama 3.2 Série | 8.192 tokens | Tarefas gerais |
| Mixtral 8x7B | 32.768 tokens | Equilíbrio entre performance e contexto |
A Groq diferencia modelos production e preview e desativa IDs regularmente. Consulte o [catálogo de modelos da Groq](https://console.groq.com/docs/models) e a [página de descontinuações](https://console.groq.com/docs/deprecations) antes de escolher um modelo para produção.
**Nota:** Este provedor usa o LiteLLM. Adicione-o como dependência ao seu projeto:
```bash
@@ -769,11 +600,12 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
Exemplo de uso em seu projeto CrewAI:
```python Code
llm = LLM(
model="llama-3.1-sonar-large-128k-online",
base_url="https://api.perplexity.ai/"
model="perplexity/sonar-pro"
)
```
Consulte o [catálogo de modelos da Perplexity](https://docs.perplexity.ai/getting-started/models) e o [changelog](https://docs.perplexity.ai/docs/resources/changelog) para obter IDs atuais e avisos de descontinuação.
**Nota:** Este provedor usa o LiteLLM. Adicione-o como dependência ao seu projeto:
```bash
uv add 'crewai[litellm]'
@@ -809,17 +641,12 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
Exemplo de uso em seu projeto CrewAI:
```python Code
llm = LLM(
model="sambanova/Meta-Llama-3.1-8B-Instruct",
model="sambanova/Meta-Llama-3.3-70B-Instruct",
temperature=0.7
)
```
| Modelo | Janela de Contexto | Melhor Para |
|-------------------|---------------------------|----------------------------------------------|
| Llama 3.1 70B/8B | Até 131.072 tokens | Alto desempenho, tarefas com grande contexto |
| Llama 3.1 405B | 8.192 tokens | Desempenho e qualidade de saída elevada |
| Llama 3.2 Série | 8.192 tokens | Tarefas gerais e multimodais |
| Llama 3.3 70B | Até 131.072 tokens | Desempenho e qualidade de saída elevada |
| Família Qwen2 | 8.192 tokens | Desempenho e qualidade de saída elevada |
Os modelos hospedados no SambaNova Cloud podem mudar independentemente do CrewAI. Consulte o [endpoint de modelos](https://docs.sambanova.ai/docs/api-reference/models/get-environments-available-model-list-metadata) e o [guia de descontinuação](https://docs.sambanova.ai/docs/en/models/deprecations) antes de implantar.
**Nota:** Este provedor usa o LiteLLM. Adicione-o como dependência ao seu projeto:
```bash
@@ -837,7 +664,7 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
Exemplo de uso em seu projeto CrewAI:
```python Code
llm = LLM(
model="cerebras/llama3.1-70b",
model="cerebras/gpt-oss-120b",
temperature=0.7,
max_tokens=8192
)
@@ -851,6 +678,8 @@ Nesta seção, você encontrará exemplos detalhados que ajudam a selecionar, co
- Suporte a longas janelas de contexto
</Info>
Consulte o [catálogo de modelos Cerebras](https://inference-docs.cerebras.ai/models/overview) e os [avisos de descontinuação](https://inference-docs.cerebras.ai/support/deprecation) para obter os IDs atuais dos endpoints públicos.
**Nota:** Este provedor usa o LiteLLM. Adicione-o como dependência ao seu projeto:
```bash
uv add 'crewai[litellm]'
@@ -898,7 +727,7 @@ O CrewAI suporta respostas em streaming de LLMs, permitindo que sua aplicação
# Crie um LLM com streaming ativado
llm = LLM(
model="openai/gpt-4o",
model="openai/gpt-5.6-terra",
stream=True # Ativar streaming
)
```
@@ -935,6 +764,8 @@ O CrewAI suporta respostas em streaming de LLMs, permitindo que sua aplicação
O CrewAI suporta respostas estruturadas de LLMs permitindo que você defina um `response_format` usando um modelo Pydantic. Isso permite que o framework automaticamente faça o parsing e valide a saída, facilitando a integração da resposta em sua aplicação sem pós-processamento manual.
O suporte a saídas estruturadas varia de acordo com o provedor e o modelo. Teste o modelo escolhido antes de depender de respostas estruturadas em produção.
Por exemplo, é possível definir um modelo Pydantic para representar a resposta esperada e passá-lo como `response_format` ao instanciar o LLM. O modelo será utilizado para converter a resposta do LLM em um objeto Python estruturado.
```python Code
@@ -946,7 +777,7 @@ class Dog(BaseModel):
breed: str
llm = LLM(model="gpt-4o", response_format=Dog)
llm = LLM(model="openai/gpt-5.6-terra", response_format=Dog)
response = llm.call(
"Analyze the following messages and return the name, age, and breed. "
@@ -975,8 +806,8 @@ Saiba como obter o máximo da configuração do seu LLM:
# 3. Divisão de tarefas para grandes contextos
llm = LLM(
model="gpt-4",
max_tokens=4000, # Limitar tamanho da resposta
model="openai/gpt-5.6-terra",
max_completion_tokens=4000, # Limitar tamanho da resposta
)
```
@@ -1000,15 +831,14 @@ Saiba como obter o máximo da configuração do seu LLM:
```python
# Configure o modelo com as opções certas
llm = LLM(
model="openai/gpt-4-turbo-preview",
temperature=0.7, # Ajuste conforme a tarefa
max_tokens=4096, # Defina conforme a necessidade da saída
timeout=300 # Timeout maior para tarefas complexas
model="openai/gpt-5.6-terra",
reasoning_effort="medium",
max_completion_tokens=4096,
timeout=300
)
```
<Tip>
- Temperaturas baixas (0.1 a 0.3) para respostas factuais
- Temperaturas altas (0.7 a 0.9) para tarefas criativas
Use os controles compatíveis com o modelo escolhido. Dependendo do provedor, isso pode ser `temperature`, um nível de reasoning ou thinking, ou instruções no prompt que definam o estilo e a variabilidade desejados.
</Tip>
</Step>