logdetective 0.2.7__tar.gz → 0.2.8__tar.gz

{logdetective-0.2.7 → logdetective-0.2.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: logdetective
-Version: 0.2.7
+Version: 0.2.8
 Summary: Log using LLM AI to search for build/test failures and provide ideas for fixing these.
 License: Apache-2.0
 Author: Jiri Podivin
@@ -14,15 +14,16 @@ Classifier: Natural Language :: English
 Classifier: Programming Language :: Python :: 3
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Internet :: Log Analysis
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Classifier: Topic :: Software Development :: Debuggers
 Provides-Extra: server
 Requires-Dist: drain3 (>=0.9.11,<0.10.0)
 Requires-Dist: huggingface-hub (>0.23.2)
-Requires-Dist: llama-cpp-python (>=0.2.56,<0.3.0,!=0.2.86)
+Requires-Dist: llama-cpp-python (>0.2.56,!=0.2.86)
 Requires-Dist: numpy (>=1.26.0,<2.0.0)
-Requires-Dist: requests (>=2.31.0,<3.0.0)
+Requires-Dist: requests (>0.2.31)
 Project-URL: homepage, https://github.com/fedora-copr/logdetective
 Project-URL: issues, https://github.com/fedora-copr/logdetective/issues
 Description-Content-Type: text/markdown
@@ -194,6 +195,11 @@ Requests can then be made with post requests, for example:
 
     curl --header "Content-Type: application/json" --request POST --data '{"url":"<YOUR_URL_HERE>"}' http://localhost:8080/analyze
 
+For more accurate responses, you can use `/analyze/staged` endpoint. This will submit snippets to model for individual analysis first.
+Afterwards the model outputs are used to construct final prompt. This will take substantially longer, compared to plain `/analyze`
+
+    curl --header "Content-Type: application/json" --request POST --data '{"url":"<YOUR_URL_HERE>"}' http://localhost:8080/analyze/staged
+
 We also have a Containerfile and composefile to run the logdetective server and llama server in containers.
 
 Before doing `podman-compose up`, make sure to set `MODELS_PATH` environment variable and point to a directory with your local model files:
@@ -205,6 +211,11 @@ $ ll $MODELS_PATH
 
 If the variable is not set, `./models` is mounted inside by default.
 
+Model can be downloaded from [our Hugging Space](https://huggingface.co/fedora-copr) by:
+```
+$ curl -L -o models/mistral-7b-instruct-v0.2.Q4_K_S.gguf https://huggingface.co/fedora-copr/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/ggml-model-Q4_K_S.gguf
+```
+
 
 License
 -------

{logdetective-0.2.7 → logdetective-0.2.8}/README.md

@@ -165,6 +165,11 @@ Requests can then be made with post requests, for example:
 
     curl --header "Content-Type: application/json" --request POST --data '{"url":"<YOUR_URL_HERE>"}' http://localhost:8080/analyze
 
+For more accurate responses, you can use `/analyze/staged` endpoint. This will submit snippets to model for individual analysis first.
+Afterwards the model outputs are used to construct final prompt. This will take substantially longer, compared to plain `/analyze`
+
+    curl --header "Content-Type: application/json" --request POST --data '{"url":"<YOUR_URL_HERE>"}' http://localhost:8080/analyze/staged
+
 We also have a Containerfile and composefile to run the logdetective server and llama server in containers.
 
 Before doing `podman-compose up`, make sure to set `MODELS_PATH` environment variable and point to a directory with your local model files:
@@ -176,6 +181,11 @@ $ ll $MODELS_PATH
 
 If the variable is not set, `./models` is mounted inside by default.
 
+Model can be downloaded from [our Hugging Space](https://huggingface.co/fedora-copr) by:
+```
+$ curl -L -o models/mistral-7b-instruct-v0.2.Q4_K_S.gguf https://huggingface.co/fedora-copr/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/ggml-model-Q4_K_S.gguf
+```
+
 
 License
 -------

{logdetective-0.2.7 → logdetective-0.2.8}/logdetective/server.py

@@ -1,12 +1,13 @@
+import asyncio
 import json
 import logging
 import os
-from typing import List
+from typing import List, Annotated
 
 from llama_cpp import CreateCompletionResponse
-from fastapi import FastAPI, HTTPException
+from fastapi import FastAPI, HTTPException, Depends, Header
+from fastapi.responses import StreamingResponse
 from pydantic import BaseModel
-
 import requests
 
 from logdetective.constants import PROMPT_TEMPLATE, SNIPPET_PROMPT_TEMPLATE
@@ -44,13 +45,42 @@ class StagedResponse(Response):
 
 LOG = logging.getLogger("logdetective")
 
-app = FastAPI()
 
 LLM_CPP_HOST = os.environ.get("LLAMA_CPP_HOST", "localhost")
 LLM_CPP_SERVER_ADDRESS = f"http://{LLM_CPP_HOST}"
 LLM_CPP_SERVER_PORT = os.environ.get("LLAMA_CPP_SERVER_PORT", 8000)
 LLM_CPP_SERVER_TIMEOUT = os.environ.get("LLAMA_CPP_SERVER_TIMEOUT", 600)
 LOG_SOURCE_REQUEST_TIMEOUT = os.environ.get("LOG_SOURCE_REQUEST_TIMEOUT", 60)
+API_TOKEN = os.environ.get("LOGDETECTIVE_TOKEN", None)
+
+def requires_token_when_set(authentication: Annotated[str | None, Header()] = None):
+    """
+    FastAPI Depend function that expects a header named Authentication
+
+    If LOGDETECTIVE_TOKEN env var is set, validate the client-supplied token
+    otherwise ignore it
+    """
+    if not API_TOKEN:
+        LOG.info("LOGDETECTIVE_TOKEN env var not set, authentication disabled")
+        # no token required, means local dev environment
+        return
+    token = None
+    if authentication:
+        try:
+            token = authentication.split(" ", 1)[1]
+        except (ValueError, IndexError):
+            LOG.warning(
+                "Authentication header has invalid structure (%s), it should be 'Bearer TOKEN'",
+                authentication)
+            # eat the exception and raise 401 below
+            token = None
+        if token == API_TOKEN:
+            return
+    LOG.info("LOGDETECTIVE_TOKEN env var is set (%s), clien token = %s",
+             API_TOKEN, token)
+    raise HTTPException(status_code=401, detail=f"Token {token} not valid.")
+
+app = FastAPI(dependencies=[Depends(requires_token_when_set)])
 
 
 def process_url(url: str) -> str:
@@ -91,7 +121,8 @@ def mine_logs(log: str) -> List[str]:
 
     return log_summary
 
-def submit_text(text: str, max_tokens: int = 0, log_probs: int = 1):
+async def submit_text(text: str, max_tokens: int = 0, log_probs: int = 1, stream: bool = False,
+                      model: str = "default-model"):
     """Submit prompt to LLM.
     max_tokens: number of tokens to be produces, 0 indicates run until encountering EOS
     log_probs: number of token choices to produce log probs for
@@ -100,7 +131,9 @@ def submit_text(text: str, max_tokens: int = 0, log_probs: int = 1):
     data = {
             "prompt": text,
             "max_tokens": str(max_tokens),
-            "logprobs": str(log_probs)}
+            "logprobs": str(log_probs),
+            "stream": stream,
+            "model": model}
 
     try:
         # Expects llama-cpp server to run on LLM_CPP_SERVER_ADDRESS:LLM_CPP_SERVER_PORT
@@ -108,24 +141,27 @@ def submit_text(text: str, max_tokens: int = 0, log_probs: int = 1):
             f"{LLM_CPP_SERVER_ADDRESS}:{LLM_CPP_SERVER_PORT}/v1/completions",
             headers={"Content-Type":"application/json"},
             data=json.dumps(data),
-            timeout=int(LLM_CPP_SERVER_TIMEOUT))
+            timeout=int(LLM_CPP_SERVER_TIMEOUT),
+            stream=stream)
     except requests.RequestException as ex:
         raise HTTPException(
             status_code=400,
             detail=f"Llama-cpp query failed: {ex}") from ex
-
-    if not response.ok:
-        raise HTTPException(
-            status_code=400,
-            detail="Something went wrong while getting a response from the llama server: "
-                f"[{response.status_code}] {response.text}")
-    try:
-        response = json.loads(response.text)
-    except UnicodeDecodeError as ex:
-        LOG.error("Error encountered while parsing llama server response: %s", ex)
-        raise HTTPException(
-            status_code=400,
-            detail=f"Couldn't parse the response.\nError: {ex}\nData: {response.text}") from ex
+    if not stream:
+        if not response.ok:
+            raise HTTPException(
+                status_code=400,
+                detail="Something went wrong while getting a response from the llama server: "
+                    f"[{response.status_code}] {response.text}")
+        try:
+            response = json.loads(response.text)
+        except UnicodeDecodeError as ex:
+            LOG.error("Error encountered while parsing llama server response: %s", ex)
+            raise HTTPException(
+                status_code=400,
+                detail=f"Couldn't parse the response.\nError: {ex}\nData: {response.text}") from ex
+    else:
+        return response
 
     return CreateCompletionResponse(response)
 
@@ -140,7 +176,8 @@ async def analyze_log(build_log: BuildLog):
     """
     log_text = process_url(build_log.url)
     log_summary = mine_logs(log_text)
-    response = submit_text(PROMPT_TEMPLATE.format(log_summary))
+    response = await submit_text(PROMPT_TEMPLATE.format(log_summary))
+    certainty = 0
 
     if "logprobs" in response["choices"][0]:
         try:
@@ -167,13 +204,11 @@ async def analyze_log_staged(build_log: BuildLog):
     log_text = process_url(build_log.url)
     log_summary = mine_logs(log_text)
 
-    analyzed_snippets = []
+    # Process snippets asynchronously
+    analyzed_snippets = await asyncio.gather(
+        *[submit_text(SNIPPET_PROMPT_TEMPLATE.format(s)) for s in log_summary])
 
-    for snippet in log_summary:
-        response = submit_text(SNIPPET_PROMPT_TEMPLATE.format(snippet))
-        analyzed_snippets.append(response)
-
-    final_analysis = submit_text(
+    final_analysis = await submit_text(
         PROMPT_TEMPLATE.format([e["choices"][0]["text"] for e in analyzed_snippets]))
 
     certainty = 0
@@ -190,3 +225,18 @@ async def analyze_log_staged(build_log: BuildLog):
 
     return StagedResponse(
         explanation=final_analysis, snippets=analyzed_snippets, response_certainty=certainty)
+
+
+@app.post("/analyze/stream", response_class=StreamingResponse)
+async def analyze_log_stream(build_log: BuildLog):
+    """Stream response endpoint for Logdetective.
+    Request must be in form {"url":"<YOUR_URL_HERE>"}.
+    URL must be valid for the request to be passed to the LLM server.
+    Meaning that it must contain appropriate scheme, path and netloc,
+    while lacking  result, params or query fields.
+    """
+    log_text = process_url(build_log.url)
+    log_summary = mine_logs(log_text)
+    stream = await submit_text(PROMPT_TEMPLATE.format(log_summary), stream=True)
+
+    return StreamingResponse(stream)

{logdetective-0.2.7 → logdetective-0.2.8}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "logdetective"
-version = "0.2.7"
+version = "0.2.8"
 description = "Log using LLM AI to search for build/test failures and provide ideas for fixing these."
 authors = ["Jiri Podivin <jpodivin@gmail.com>"]
 license = "Apache-2.0"
@@ -27,8 +27,8 @@ issues = "https://github.com/fedora-copr/logdetective/issues"
 
 [tool.poetry.dependencies]
 python = "^3.11"
-requests = "^2.31.0"
-llama-cpp-python = "^0.2.56,!=0.2.86"
+requests = ">0.2.31"
+llama-cpp-python = ">0.2.56,!=0.2.86"
 drain3 = "^0.9.11"
 huggingface-hub = ">0.23.2"
 numpy = "^1.26.0"