Merge pull request unclecode#1440 from unclecode/feature/docker-llm-parameters

feat(docker): Add temperature and base_url parameters for LLM configuration

Author: ntohidi

deploy/docker/.llm.env.example

@@ -10,4 +10,23 @@ GEMINI_API_TOKEN=your_gemini_key_here
 # Optional: Override the default LLM provider
 # Examples: "openai/gpt-4", "anthropic/claude-3-opus", "deepseek/chat", etc.
 # If not set, uses the provider specified in config.yml (default: openai/gpt-4o-mini)
-# LLM_PROVIDER=anthropic/claude-3-opus
+# LLM_PROVIDER=anthropic/claude-3-opus
+
+# Optional: Global LLM temperature setting (0.0-2.0)
+# Controls randomness in responses. Lower = more focused, Higher = more creative
+# LLM_TEMPERATURE=0.7
+
+# Optional: Global custom API base URL
+# Use this to point to custom endpoints or proxy servers
+# LLM_BASE_URL=https://api.custom.com/v1
+
+# Optional: Provider-specific temperature overrides
+# These take precedence over the global LLM_TEMPERATURE
+# OPENAI_TEMPERATURE=0.5
+# ANTHROPIC_TEMPERATURE=0.3
+# GROQ_TEMPERATURE=0.8
+
+# Optional: Provider-specific base URL overrides
+# Use for provider-specific proxy endpoints
+# OPENAI_BASE_URL=https://custom-openai.company.com/v1
+# GROQ_BASE_URL=https://custom-groq.company.com/v1

deploy/docker/api.py

@@ -42,7 +42,9 @@
     should_cleanup_task,
     decode_redis_hash,
     get_llm_api_key,
-    validate_llm_provider
+    validate_llm_provider,
+    get_llm_temperature,
+    get_llm_base_url
 )
 
 import psutil, time
@@ -96,7 +98,9 @@ async def handle_llm_qa(
         response = perform_completion_with_backoff(
             provider=config["llm"]["provider"],
             prompt_with_variables=prompt,
-            api_token=get_llm_api_key(config)  # Returns None to let litellm handle it
+            api_token=get_llm_api_key(config),  # Returns None to let litellm handle it
+            temperature=get_llm_temperature(config),
+            base_url=get_llm_base_url(config)
         )
 
         return response.choices[0].message.content
@@ -115,7 +119,9 @@ async def process_llm_extraction(
     instruction: str,
     schema: Optional[str] = None,
     cache: str = "0",
-    provider: Optional[str] = None
+    provider: Optional[str] = None,
+    temperature: Optional[float] = None,
+    base_url: Optional[str] = None
 ) -> None:
     """Process LLM extraction in background."""
     try:
@@ -131,7 +137,9 @@ async def process_llm_extraction(
         llm_strategy = LLMExtractionStrategy(
             llm_config=LLMConfig(
                 provider=provider or config["llm"]["provider"],
-                api_token=api_key
+                api_token=api_key,
+                temperature=temperature or get_llm_temperature(config, provider),
+                base_url=base_url or get_llm_base_url(config, provider)
             ),
             instruction=instruction,
             schema=json.loads(schema) if schema else None,
@@ -178,7 +186,9 @@ async def handle_markdown_request(
     query: Optional[str] = None,
     cache: str = "0",
     config: Optional[dict] = None,
-    provider: Optional[str] = None
+    provider: Optional[str] = None,
+    temperature: Optional[float] = None,
+    base_url: Optional[str] = None
 ) -> str:
     """Handle markdown generation requests."""
     try:
@@ -204,6 +214,8 @@ async def handle_markdown_request(
                     llm_config=LLMConfig(
                         provider=provider or config["llm"]["provider"],
                         api_token=get_llm_api_key(config, provider),  # Returns None to let litellm handle it
+                        temperature=temperature or get_llm_temperature(config, provider),
+                        base_url=base_url or get_llm_base_url(config, provider)
                     ),
                     instruction=query or "Extract main content"
                 )
@@ -248,7 +260,9 @@ async def handle_llm_request(
     schema: Optional[str] = None,
     cache: str = "0",
     config: Optional[dict] = None,
-    provider: Optional[str] = None
+    provider: Optional[str] = None,
+    temperature: Optional[float] = None,
+    api_base_url: Optional[str] = None
 ) -> JSONResponse:
     """Handle LLM extraction requests."""
     base_url = get_base_url(request)
@@ -279,7 +293,9 @@ async def handle_llm_request(
             cache,
             base_url,
             config,
-            provider
+            provider,
+            temperature,
+            api_base_url
         )
 
     except Exception as e:
@@ -324,7 +340,9 @@ async def create_new_task(
     cache: str,
     base_url: str,
     config: dict,
-    provider: Optional[str] = None
+    provider: Optional[str] = None,
+    temperature: Optional[float] = None,
+    api_base_url: Optional[str] = None
 ) -> JSONResponse:
     """Create and initialize a new task."""
     decoded_url = unquote(input_path)
@@ -349,7 +367,9 @@ async def create_new_task(
         query,
         schema,
         cache,
-        provider
+        provider,
+        temperature,
+        api_base_url
     )
 
     return JSONResponse({

deploy/docker/job.py

@@ -37,6 +37,8 @@ class LlmJobPayload(BaseModel):
     schema: Optional[str] = None
     cache:  bool = False
     provider: Optional[str] = None
+    temperature: Optional[float] = None
+    base_url: Optional[str] = None
 
 
 class CrawlJobPayload(BaseModel):
@@ -63,6 +65,8 @@ async def llm_job_enqueue(
         cache=payload.cache,
         config=_config,
         provider=payload.provider,
+        temperature=payload.temperature,
+        api_base_url=payload.base_url,
     )
 
 
@@ -72,7 +76,7 @@ async def llm_job_status(
     task_id: str,
     _td: Dict = Depends(lambda: _token_dep())
 ):
-    return await handle_task_status(_redis, task_id)
+    return await handle_task_status(_redis, task_id, base_url=str(request.base_url))
 
 
 # ---------- CRAWL job -------------------------------------------------------

deploy/docker/schemas.py

@@ -16,6 +16,8 @@ class MarkdownRequest(BaseModel):
     q:   Optional[str] = Field(None,  description="Query string used by BM25/LLM filters")
     c:   Optional[str] = Field("0",   description="Cache‑bust / revision counter")
     provider: Optional[str] = Field(None, description="LLM provider override (e.g., 'anthropic/claude-3-opus')")
+    temperature: Optional[float] = Field(None, description="LLM temperature override (0.0-2.0)")
+    base_url: Optional[str] = Field(None, description="LLM API base URL override")
 
 
 class RawCode(BaseModel):

deploy/docker/server.py

@@ -241,7 +241,8 @@ async def get_markdown(
         raise HTTPException(
             400, "Invalid URL format. Must start with http://, https://, or for raw HTML (raw:, raw://)")
     markdown = await handle_markdown_request(
-        body.url, body.f, body.q, body.c, config, body.provider
+        body.url, body.f, body.q, body.c, config, body.provider,
+        body.temperature, body.base_url
     )
     return JSONResponse({
         "url": body.url,

deploy/docker/utils.py

@@ -108,6 +108,69 @@ def validate_llm_provider(config: Dict, provider: Optional[str] = None) -> tuple
     return True, ""
 
 
+def get_llm_temperature(config: Dict, provider: Optional[str] = None) -> Optional[float]:
+    """Get temperature setting based on the LLM provider.
+    
+    Priority order:
+    1. Provider-specific environment variable (e.g., OPENAI_TEMPERATURE)
+    2. Global LLM_TEMPERATURE environment variable
+    3. None (to use litellm/provider defaults)
+    
+    Args:
+        config: The application configuration dictionary
+        provider: Optional provider override (e.g., "openai/gpt-4")
+    
+    Returns:
+        The temperature setting if configured, otherwise None
+    """
+    # Check provider-specific temperature first
+    if provider:
+        provider_name = provider.split('/')[0].upper()
+        provider_temp = os.environ.get(f"{provider_name}_TEMPERATURE")
+        if provider_temp:
+            try:
+                return float(provider_temp)
+            except ValueError:
+                logging.warning(f"Invalid temperature value for {provider_name}: {provider_temp}")
+    
+    # Check global LLM_TEMPERATURE
+    global_temp = os.environ.get("LLM_TEMPERATURE")
+    if global_temp:
+        try:
+            return float(global_temp)
+        except ValueError:
+            logging.warning(f"Invalid global temperature value: {global_temp}")
+    
+    # Return None to use litellm/provider defaults
+    return None
+
+
+def get_llm_base_url(config: Dict, provider: Optional[str] = None) -> Optional[str]:
+    """Get base URL setting based on the LLM provider.
+    
+    Priority order:
+    1. Provider-specific environment variable (e.g., OPENAI_BASE_URL)
+    2. Global LLM_BASE_URL environment variable
+    3. None (to use default endpoints)
+    
+    Args:
+        config: The application configuration dictionary
+        provider: Optional provider override (e.g., "openai/gpt-4")
+    
+    Returns:
+        The base URL if configured, otherwise None
+    """
+    # Check provider-specific base URL first
+    if provider:
+        provider_name = provider.split('/')[0].upper()
+        provider_url = os.environ.get(f"{provider_name}_BASE_URL")
+        if provider_url:
+            return provider_url
+    
+    # Check global LLM_BASE_URL
+    return os.environ.get("LLM_BASE_URL")
+
+
 def verify_email_domain(email: str) -> bool:
     try:
         domain = email.split('@')[1]