auto-coder 0.1.203__tar.gz → 0.1.204__tar.gz

{auto_coder-0.1.203 → auto_coder-0.1.204}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: auto-coder
-Version: 0.1.203
+Version: 0.1.204
 Summary: AutoCoder: AutoCoder
 Author: allwefantasy
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/auto_coder.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: auto-coder
-Version: 0.1.203
+Version: 0.1.204
 Summary: AutoCoder: AutoCoder
 Author: allwefantasy
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/autocoder/auto_coder_rag.py

@@ -286,6 +286,24 @@ def main(input_args: Optional[List[str]] = None):
         help="Whether to return responses without contexts. only works when pro plugin is installed",
     )
 
+    serve_parser.add_argument(
+        "--recall_model",
+        default="",
+        help="The model used for recall documents",
+    )
+
+    serve_parser.add_argument(
+        "--chunk_model",
+        default="",
+        help="The model used for chunk documents",
+    )
+
+    serve_parser.add_argument(
+        "--qa_model",
+        default="",
+        help="The model used for question answering",
+    )
+
     # Tools command
     tools_parser = subparsers.add_parser("tools", help="Various tools")
     tools_subparsers = tools_parser.add_subparsers(dest="tool", help="Available tools")
@@ -334,6 +352,22 @@ def main(input_args: Optional[List[str]] = None):
         llm = byzerllm.ByzerLLM()
         llm.setup_default_model_name(args.model)
 
+        # Setup sub models if specified
+        if args.recall_model:
+            recall_model = byzerllm.ByzerLLM()
+            recall_model.setup_default_model_name(args.recall_model)
+            llm.setup_sub_client("recall_model", recall_model)
+
+        if args.chunk_model:
+            chunk_model = byzerllm.ByzerLLM()
+            chunk_model.setup_default_model_name(args.chunk_model)
+            llm.setup_sub_client("chunk_model", chunk_model)
+
+        if args.qa_model:
+            qa_model = byzerllm.ByzerLLM()
+            qa_model.setup_default_model_name(args.qa_model)
+            llm.setup_sub_client("qa_model", qa_model)
+
         # 当启用hybrid_index时,检查必要的组件
         if auto_coder_args.enable_hybrid_index:                           
             if not llm.is_model_exist("emb"):

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/autocoder/common/code_auto_generate.py

@@ -172,7 +172,7 @@ class CodeAutoGenerate:
         conversations = []
 
         if self.args.system_prompt and self.args.system_prompt.strip() == "claude":
-            conversations.append({"role": "system", "content": sys_prompt.prompt()})
+            conversations.append({"role": "system", "content": sys_prompt.claude_sys_prompt.prompt()})
         elif self.args.system_prompt:
             conversations.append({"role": "system", "content": self.args.system_prompt})
         

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/autocoder/common/code_auto_generate_diff.py

@@ -308,7 +308,7 @@ class CodeAutoGenerateDiff:
 
         conversations = []
         if self.args.system_prompt and self.args.system_prompt.strip() == "claude":
-            conversations.append({"role": "system", "content": sys_prompt.prompt()})
+            conversations.append({"role": "system", "content": sys_prompt.claude_sys_prompt.prompt()})
         else:
             conversations.append({"role": "user", "content": init_prompt})
 

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/autocoder/common/code_auto_generate_editblock.py

@@ -390,7 +390,7 @@ class CodeAutoGenerateEditBlock:
         conversations = []
         
         if self.args.system_prompt and self.args.system_prompt.strip() == "claude":
-            conversations.append({"role": "system", "content": sys_prompt.prompt()})
+            conversations.append({"role": "system", "content": sys_prompt.claude_sys_prompt.prompt()})
         elif self.args.system_prompt:
             conversations.append({"role": "system", "content": self.args.system_prompt})
         

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/autocoder/common/code_auto_generate_strict_diff.py

@@ -279,7 +279,7 @@ class CodeAutoGenerateStrictDiff:
 
         conversations = []
         if self.args.system_prompt and self.args.system_prompt.strip() == "claude":
-            conversations.append({"role": "system", "content": sys_prompt.prompt()})
+            conversations.append({"role": "system", "content": sys_prompt.claude_sys_prompt.prompt()})
         elif self.args.system_prompt:
             conversations.append({"role": "system", "content": self.args.system_prompt})
         

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/autocoder/common/sys_prompt.py

@@ -1,5 +1,5 @@
 import byzerllm
-import datetime
+from datetime import datetime
 
 @byzerllm.prompt()
 def claude_sys_prompt():

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/autocoder/rag/doc_filter.py

@@ -62,6 +62,10 @@ def _check_relevance_with_conversation(
 class DocFilterWorker:
     def __init__(self, llm: ByzerLLM):
         self.llm = llm
+        if self.llm.get_sub_client("recall_model"):
+            self.recall_llm = self.llm.get_sub_client("recall_model")
+        else:
+            self.recall_llm = self.llm
 
     def filter_doc(
         self, conversations: List[Dict[str, str]], docs: List[str]
@@ -72,7 +76,8 @@ class DocFilterWorker:
                 conversations=conversations, documents=docs
             )
         except Exception as e:
-            logger.error(f"Error in _check_relevance_with_conversation: {str(e)}")
+            logger.error(
+                f"Error in _check_relevance_with_conversation: {str(e)}")
             return (None, submit_time_1, time.time())
 
         end_time_2 = time.time()
@@ -88,6 +93,11 @@ class DocFilter:
         path: Optional[str] = None,
     ):
         self.llm = llm
+        if self.llm.get_sub_client("recall_model"):
+            self.recall_llm = self.llm.get_sub_client("recall_model")
+        else:
+            self.recall_llm = self.llm
+        
         self.args = args
         self.relevant_score = self.args.rag_doc_filter_relevance or 5
         self.on_ray = on_ray
@@ -95,7 +105,8 @@ class DocFilter:
         if self.on_ray:
             cpu_count = os.cpu_count() or 1
             self.workers = [
-                DocFilterWorker.options(max_concurrency=1000, num_cpus=0).remote(llm)
+                DocFilterWorker.options(
+                    max_concurrency=1000, num_cpus=0).remote(llm)
                 for _ in range(cpu_count)
             ]
 
@@ -137,10 +148,12 @@ class DocFilter:
                         submit_time_1 = time.time()
                         try:
                             llm = ByzerLLM()
-                            llm.setup_default_model_name(self.llm.default_model_name)
                             llm.skip_nontext_check = True
+                            llm.setup_default_model_name(self.recall_llm.default_model_name)
+
                             v = (
-                                _check_relevance_with_conversation.with_llm(llm)
+                                _check_relevance_with_conversation.with_llm(
+                                    llm)
                                 .options({"llm_config": {"max_length": 10}})
                                 .run(
                                     conversations=conversations,
@@ -194,10 +207,12 @@ class DocFilter:
                             )
                         )
                 except Exception as exc:
-                    logger.error(f"Document processing generated an exception: {exc}")
+                    logger.error(
+                        f"Document processing generated an exception: {exc}")
 
         # Sort relevant_docs by relevance score in descending order
-        relevant_docs.sort(key=lambda x: x.relevance.relevant_score, reverse=True)
+        relevant_docs.sort(
+            key=lambda x: x.relevance.relevant_score, reverse=True)
         return relevant_docs
 
     def filter_docs_with_ray(
@@ -210,7 +225,8 @@ class DocFilter:
             worker = self.workers[count % len(self.workers)]
             count += 1
             future = worker.filter_doc.remote(
-                conversations, [f"##File: {doc.module_name}\n{doc.source_code}"]
+                conversations, [
+                    f"##File: {doc.module_name}\n{doc.source_code}"]
             )
             futures.append((future, doc))
 
@@ -248,8 +264,10 @@ class DocFilter:
                         )
                     )
             except Exception as exc:
-                logger.error(f"Document processing generated an exception: {exc}")
+                logger.error(
+                    f"Document processing generated an exception: {exc}")
 
         # Sort relevant_docs by relevance score in descending order
-        relevant_docs.sort(key=lambda x: x.relevance.relevant_score, reverse=True)
+        relevant_docs.sort(
+            key=lambda x: x.relevance.relevant_score, reverse=True)
         return relevant_docs

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/autocoder/rag/long_context_rag.py

@@ -341,6 +341,11 @@ class LongContextRAG:
 
             return response_generator(), []
         else:
+            
+            target_llm = self.llm
+            if self.llm.get_sub_client("qa_model"):
+                target_llm = self.llm.get_sub_client("qa_model")
+
             query = conversations[-1]["content"]
             context = []
 
@@ -349,8 +354,9 @@ class LongContextRAG:
                 in query
                 or "简要总结一下对话内容，用作后续的上下文提示 prompt，控制在 200 字以内"
                 in query
-            ):
-                chunks = self.llm.stream_chat_oai(
+            ):                
+
+                chunks = target_llm.stream_chat_oai(
                     conversations=conversations,
                     model=model,
                     role_mapping=role_mapping,
@@ -384,7 +390,7 @@ class LongContextRAG:
 
             if self.args.without_contexts and LLMComputeEngine is not None:
                 llm_compute_engine = LLMComputeEngine(
-                    llm=self.llm,
+                    llm=target_llm,
                     inference_enhance=not self.args.disable_inference_enhance,
                     inference_deep_thought=self.args.inference_deep_thought,
                     inference_slow_without_deep_thought=self.args.inference_slow_without_deep_thought,                    
@@ -566,7 +572,7 @@ class LongContextRAG:
 
             if LLMComputeEngine is not None and not self.args.disable_inference_enhance:
                 llm_compute_engine = LLMComputeEngine(
-                    llm=self.llm,
+                    llm=target_llm,
                     inference_enhance=not self.args.disable_inference_enhance,
                     inference_deep_thought=self.args.inference_deep_thought,
                     precision=self.args.inference_compute_precision,
@@ -597,7 +603,7 @@ class LongContextRAG:
                 }
             ]
 
-            chunks = self.llm.stream_chat_oai(
+            chunks = target_llm.stream_chat_oai(
                 conversations=new_conversations,
                 model=model,
                 role_mapping=role_mapping,

{auto_coder-0.1.203 → auto_coder-0.1.204}/src/autocoder/rag/token_limiter.py

@@ -16,7 +16,7 @@ class TokenLimiter:
         full_text_limit: int,
         segment_limit: int,
         buff_limit: int,
-        llm,
+        llm:ByzerLLM,
         disable_segment_reorder: bool,
     ):
         self.count_tokens = count_tokens
@@ -24,6 +24,10 @@ class TokenLimiter:
         self.segment_limit = segment_limit
         self.buff_limit = buff_limit
         self.llm = llm
+        if self.llm.get_sub_client("chunk_model"):
+            self.chunk_llm = self.llm.get_sub_client("chunk_model")
+        else:
+            self.chunk_llm = self.llm
         self.first_round_full_docs = []
         self.second_round_extracted_docs = []
         self.sencond_round_time = 0
@@ -219,10 +223,10 @@ class TokenLimiter:
                 source_code_lines = doc.source_code.split("\n")
                 for idx, line in enumerate(source_code_lines):
                     source_code_with_line_number += f"{idx+1} {line}\n"
-
+                
                 llm = ByzerLLM()
-                llm.setup_default_model_name(self.llm.default_model_name)
                 llm.skip_nontext_check = True
+                llm.setup_default_model_name(self.chunk_llm.default_model_name)
 
                 extracted_info = (
                     self.extract_relevance_range_from_docs_with_conversation.options(

auto_coder-0.1.204/src/autocoder/version.py

@@ -0,0 +1 @@
+__version__ = "0.1.204"

auto_coder-0.1.203/src/autocoder/version.py

@@ -1 +0,0 @@
-__version__ = "0.1.203"