Skip to content

DEEPSEEK_R1_CONFIG.md

Latest commit

 

History

History
95 lines (73 loc) · 2.81 KB

DEEPSEEK_R1_CONFIG.md

File metadata and controls

95 lines (73 loc) · 2.81 KB

DeepSeek R1 Configuration Summary

Problem

DeepSeek R1 has reasoning always enabled at the model level. The model:

  • Always returns reasoningContent in a separate content block (Block 0)
  • Always returns the final answer in a text block (Block 1)
  • Does NOT accept thinking or reasoning_config parameters in additionalModelRequestFields
  • Sending reasoning parameters causes validation errors

Solution

Added a new flag no_reasoning_params to the reasoning configuration object.

Configuration for DeepSeek R1

"DeepSeek R1": {
  "id": "us.deepseek.r1-v1:0",
  "inference_profile": {
    "prefix": "us",
    "region": "us-west-2"
  },
  "region": ["us-east-1", "us-east-2", "us-west-2"],
  "cost": {
    "input_1k_price": 0.00135,
    "output_1k_price": 0.0054
  },
  "maxTokens": 163840,
  "vision": false,
  "document": true,
  "tool": false,
  "streaming": true,
  "reasoning": {
    "enabled": true,
    "openai_reasoning_modalities": true,
    "no_reasoning_params": true
  }
}

What Each Flag Does

  • enabled: true: Tells the app that reasoning is supported
  • openai_reasoning_modalities: true:
    • Look for reasoningContent field in the response (not thinking)
    • Show reasoning as always-on in the UI (no toggle)
    • Handle two-block response structure
  • no_reasoning_params: true:
    • NEW FLAG - Prevents sending ANY reasoning parameters to the API
    • No additionalModelRequestFields for reasoning
    • Avoids validation errors from the model

Code Changes

1. TypeScript Interface (bin/config.ts)

Added no_reasoning_params?: boolean to ReasoningConfig interface with documentation.

2. Python App Logic (chainlit_image/foundational-llm-chat_app/app.py)

Added check before sending reasoning parameters:

if isinstance(reasoning_config, dict) and reasoning_config.get("no_reasoning_params"):
    logger.debug("Model has reasoning always enabled - not sending any reasoning parameters")
    # Don't add any reasoning parameters to additional_model_fields
elif isinstance(reasoning_config, dict) and reasoning_config.get("openai_reasoning_modalities"):
    # ... existing OpenAI logic
else:
    # ... existing Anthropic logic

3. Documentation (README.md)

  • Added no_reasoning_params to the configuration parameter list
  • Added new example section "Model-Level Reasoning (Always On)"
  • Explained when and why to use this flag

Testing

Verified that DeepSeek R1:

  • ✅ Works without any additionalModelRequestFields
  • ✅ Returns reasoningContent in Block 0
  • ✅ Returns final answer in Block 1
  • ❌ Fails when sending thinking parameter
  • ✅ Works with the new configuration

Impact

  • No breaking changes to existing models
  • Only affects models with no_reasoning_params: true
  • All other reasoning configurations work exactly as before