info/mistral-api
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

integrated sentence transformer model

Browse Source
This commit is contained in:
admin 2025-12-08 11:07:18 +01:00
parent c7c7162311
commit ddc1c56cd7
2 changed files with 39 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
Dockerfile
View File

@ -53,7 +53,15 @@ mapping = {
choice = os.environ.get("RAG_EMBEDDINGS","gte-multilingual").lower()
hf_id = mapping.get(choice, "BAAI/bge-small-en-v1.5")
# cache_folder respecteert SENTENCE_TRANSFORMERS_HOME/HF_HOME, maar we forceren expliciet:
SentenceTransformer(hf_id, cache_folder=os.environ.get("SENTENCE_TRANSFORMERS_HOME","/opt/sentence-transformers"))
cache_root = os.environ.get("SENTENCE_TRANSFORMERS_HOME", "/opt/sentence-transformers")
local_dir = os.path.join(cache_root, "embedder")
os.makedirs(cache_root, exist_ok = True)
print("Downloading SentenceTransformer:", hf_id)
# Laat SentenceTransformer zelf hun cache doen (HF_HOME etc)
# wij saven het eindresultaat daarna naar local_dir
model = SentenceTransformer(hf_id, cache_folder=os.environ.get("SENTENCE_TRANSFORMERS_HOME","/opt/sentence-transformers"))
model.save(local_dir)
print("Prefetched SentenceTransformer:", hf_id)
PY

35
app.py
View File

@ -190,10 +190,21 @@ def _extract_tool_calls_from_text(txt: str):
raw = obj["call_tool"]
if isinstance(raw, dict):
raw = [raw]
if isinstance(raw, str):
print("Invalid toolcall:",str(raw))
raw=[]
tc = []
for it in (raw or []):
name = (it or {}).get("name")
args = (it or {}).get("arguments") or {}
try:
name = (it or {}).get("name")
except Exception as e:
print("Error:",str(e),"raw=",str(raw),"it=",str(it))
continue
try:
args = (it or {}).get("arguments") or {}
except Exception as e:
print("Error:",str(e),"raw=",str(raw),"it=",str(it))
continue
if isinstance(args, str):
try: args = json.loads(args)
except Exception: pass
@ -540,11 +551,23 @@ def _build_embedder() -> _Embedder:
model_name, family, slug = mapping["bge-small"]
else:
model_name, family, slug = mapping[choice]
cache_dir = os.environ.get("SENTENCE_TRANSFORMERS_HOME", "/opt/sentence-transformers")
local_dir = os.path.join(cache_dir, "embedder")
st_kwargs = {"device": "cpu"}
if os.path.isdir(local_dir):
# Prefetched model in image → gebruik dat
model_source = local_dir
print(f"Loading SentenceTransformer from local dir: {local_dir}")
else:
# Geen lokale dir gevonden → terugvallen op HF-ID + cache_folder
model_source = model_name
st_kwargs["cache_folder"] = cache_dir
print(f"Local dir {local_dir} not found, falling back to HF model: {model_name}")
try:
if "trust_remote_code" in inspect.signature(SentenceTransformer).parameters:
st_kwargs["trust_remote_code"] = True
model = SentenceTransformer(model_name, **st_kwargs)
model = SentenceTransformer(model_source, **st_kwargs)
# optioneel: CPU thread-telling forceren
try:
thr = int(os.getenv("RAG_TORCH_THREADS", "0"))
@ -554,8 +577,8 @@ def _build_embedder() -> _Embedder:
except Exception:
pass
return _Embedder(slug=slug, family=family, model=model, device="cpu")
except Exception:
pass
except Exception as e:
print("ERROR building embedder:",str(e))
# Fallback via Chroma embedding functions
from chromadb.utils import embedding_functions as ef
@ -2630,6 +2653,8 @@ async def openai_chat_completions(body: dict = Body(...), request: Request = Non
})
call = (obj.get("call_tool") or {})
if isinstance(call, str):
call={"name": "Invalid_toolcall", "arguments": {}}
tname = call.get("name")
if tname not in TOOLS_REGISTRY:
return JSONResponse(_openai_chat_response(model, f"Onbekende tool: {tname}", messages))