cmc

cmc

cmc

cmc

cmc

cmc

cmc

cmc

clean final: HF Inference API embeddings (no disk)

clean final: HF Inference API embeddings (no disk)

clean final: HF Inference API embeddings (no disk)

clean final: HF Inference API embeddings (no disk)

use HuggingFaceInferenceAPIEmbeddings (current wrapper)

drop forbidden api_url key (HF Inference default)

migrate to HF Inference Providers router (no disk, no cache)

use optimum ONNX MiniLM (no disk, no internet)

force offline local MiniLM folder (no cache)

use pre-installed MiniLM (no cache, no internet)

force offline local MiniLM (no internet, no cache)

load local MiniLM folder (no optimum, no internet)

use optimum ONNX MiniLM via low-level wrapper (no disk)

use optimum ONNX MiniLM (no disk, no internet)

load local ONNX MiniLM with absolute path (no cache)

okay

o

force local st_model folder (no internet, no cache)

add local MiniLM-L6-v2 model (384 KB, no cache, no internet)

load ST model once into RAM (no disk, no external API)

migrate to HF Inference Providers router (no disk)

use in-memory ST model (no disk, no inference API)

cmc

switch to HuggingFaceInferenceAPIEmbeddings (no disk)

use public HF inference endpoint for embeddings (no disk)

cmc

create writable /code/.cache dir

force cache into writable /code/.cache

bullet-proof fetch + empty corpus fallback

drop build cache scripts – use RAM-only rag

cmc

switch to text-only dataset for offline build

cmc

cmc

o

o

cmc

cmc

o

force 0.3+ langchain wheels

cmc

add base langchain package