Spaces:
Running
Running
Commit History
-
fix(deploy): point llama-server at the Gemma 4 GGUF (start scripts)
nmrenyi +1: Claude Opus 4.8The first Gemma 4 commit updated config.py/Dockerfile/fetch_assets.sh but missed deploy/start.sh + scripts/run.sh, which still defaulted GGUF to the (no-longer- downloaded) gemma-3n-E4B-it-Q4_0.gguf — so the Space built fine but llama-server crashed at startup (model file not found). Both now default to gemma-4-E4B-it-Q4_0.gguf. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Rb8sCfVzRDRZko7V5JYxUB4745876 -
feat: mirror device update — Gemma 4 E4B + G1 prompt
nmrenyi +1: Claude Opus 4.8Demo tracks the deployed app, which moved to Gemma 4 E4B + the G1 deflection-fix prompt (English-only). Changes: - generator: gemma-3n-E4B-it-Q4_0.gguf -> gemma-4-E4B-it-Q4_0.gguf (unsloth/gemma-4-E4B-it-GGUF, ~4.84 GB) in config.py, fetch_assets.sh, Dockerfile - prompt: deploy/system_en.txt (baked into image) -> G1 (config-v0.3.0) - GENERATOR_LABEL + README fidelity table + app.py/DEPLOYMENT.md refs -> Gemma 4 Retriever (EmbeddingGemma v0.3.0) + gen params + context injection unchanged. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Rb8sCfVzRDRZko7V5JYxUBe4ce8ab
-
chore(scripts): add asset-fetch and local run scripts
nmrenyi +1: Claude Opus 4.8 (1M context)fetch_assets.sh pulls the EmbeddingGemma retriever and Q4_0 GGUF generator and links the local v0.3.0 store + system prompt. run.sh launches llama-server (Metal by default, CPU-overridable) and uvicorn, waiting on llama-server health. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>c646f8b