Commit History
Add newer/smaller default models per role: Qwen3-4B-Thinking (reasoning), Dolphin3.0-3B (uncensored) f3aafd7
Swap Dolphin 8B for Dolphin 7B (8B was too slow for free CPU tier, regularly hit the 90s self-heal timeout) 558d2c2 verified
Sync: fix Qwen general-role model repo in shared catalogue 770e9e8 verified
Fix: remove hardcoded ENV MODEL_ID=qwen that was silently overriding SPACE_ROLE on every Space fd8d54e verified
Remove temp diagnostic endpoint (root cause found) d5498d8 verified
Temp diagnostic: add /jolly/debug-env 297eb68 verified
Fix: queue concurrent requests instead of instant 429 rejection b9dbe38 verified
Document SPACE_ROLE multi-space setup and /jolly/refresh-model 911e357 verified
Add role-based auto model assignment (SPACE_ROLE=reasoning) + /jolly/refresh-model 2fd12a2 verified
Add 90s generation timeout that kills+restarts the worker instead of holding the busy-lock forever on an abandoned/oversized request b8434c4
JollyDay Assistant commited on
Reduce max_tokens cap to 80 (~15s) — 180 tokens was still hitting a client-side timeout 9b9dc97
JollyDay Assistant commited on
Fail fast with 429 when already generating, instead of silently queuing behind it 58e8bc5
JollyDay Assistant commited on
Run inference in a dedicated child process; tighten max_tokens cap to 180 for tolerable response times without streaming 9c4a2b6
JollyDay Assistant commited on
Cap max_tokens at 400 — long CPU-only generations block the entire process, not just the request e2184e8
JollyDay Assistant commited on
Rewrite: load model directly with Llama class, drop buggy llama_cpp.server subprocess/proxy entirely 2f079cf
JollyDay Assistant commited on
Fix empty-body proxy responses: reuse a long-lived httpx client instead of closing it before streaming finishes 37cb455
JollyDay Assistant commited on
Strip hop-by-hop headers in proxy forward — fixes empty 400s on all /v1/* routes 591b4ad
JollyDay Assistant commited on
Fix health check: llama_cpp.server has no /health route, poll /v1/models instead 4d114a9
JollyDay Assistant commited on
Switch base image to Alpine — the pinned wheel is musl-linked, not glibc bc610c2
JollyDay Assistant commited on
Pin llama-cpp-python to fix build timeout (Job timeout / BUILD_ERROR) 1f1ea05
JollyDay Assistant commited on