Spaces:
Running
Running
Commit History
Today
-
Normalize inline system turns from Claude Code 2.1.317 Running
Move supported Claude Code system turns into the top-level Anthropic system field while retaining rejection for older clients. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>844b7df -
Sync upstream v6.19.35 and preserve custom addons
Preserve custom NIM/Groq rotation, request sanitization, model aliases, HF packaging, and empty-message validation while adapting to the upstream provider APIs. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>a9e27c0
15 days ago
-
Fix NIM request validation and image recovery
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>e25664f -
fix(nim): drop unsupported top_k keyword argument
80ee03e -
fix(nim): sanitize reasoning_effort, strip extra_body and empty tools, add rotator 400 auto-recovery
51f4255
17 days ago
-
fix(nim): omit reasoning_effort on disabled and retry on 400 rejection
5aba885 -
sync: upstream 8dfff832 (v6.18.5) + custom addons (KeyRotator, 401/403 rotation, sol/terra/luna, ScannerTrap, health, Py3.12 compat)
2e7b8ca
-
sync: upstream v6.16.4 (74d4ccfd) + 401/403 graceful key rotation + addons + Py3.12 compat
f7872b8
-
fix: restore missing RequestValidationError import in api/app.py
6a04f44 -
sync: upstream a2e9c47 + all addons + proactive NIM image cap (8 max) + Py3.12 compat
3cf54eb
-
feat(nim): auto-drop oldest images when NIM rejects >N image requests, retry with cap
9b42d0f -
fix(reasoning): apply PR #2013 native reasoning state fix (resolves #2004 / #2006) + Py3.12 compat
dbd348c
-
fix(py312): replace bare simplejson.loads() with json.loads() in 6 new upstream files
21d0536 -
fix(transport): accept on_event and kwargs in RotatingChatTransport._create_stream
6e8794d -
sync: upstream 9194f15 + custom addons (KeyRotator, RotatingChatTransport, sol/terra/luna, ScannerTrap, health, Py3.12 compat)
e99df76
-
sync: upstream 9e23905 with static v6.5.4 (fix hatch-vcs Docker build error) + custom addons
691dc7c -
Revert "sync: upstream 9e23905 + custom addons (KeyRotator, RotatingChatTransport, sol/terra/luna, ScannerTrap, health, Py3.12 compat)"
This reverts commit 6dc304d31c43fbee6bd54a0b59c70edfd2662d4d.9f6a4bd -
sync: upstream 9e23905 + custom addons (KeyRotator, RotatingChatTransport, sol/terra/luna, ScannerTrap, health, Py3.12 compat)
6dc304d
-
fix(rotator): restrict RELAY_URL to NVIDIA endpoints and expose model routes in /health
5d8860e -
fix(providers): match upstream _create_stream signature (4-tuple return, used_retry_kinds, extra_headers)
97e2439 -
fix(py312): fix super() calls in response_streams and failures for Python 3.12 compatibility
c43e502 -
sync upstream 81fa340 + custom addons (KeyRotator, sol/terra/luna, ScannerTrap, health, Kimi stream, Py3.12 compat)
8e438ab
-
feat: Groq multi-key rotator with 90s smart wait-and-retry and indexed env discovery
575939f -
feat: smart wait-and-retry key rotation on rate limits with 90s budget
9617ec1 -
debug: log APIConnectionError root cause
4eb8f20
-
feat: relay URL support for IP rotation
d6c95ef -
debug: log 429 body + retry-after
66b1e51 -
debug: print() tracing in key rotation loop
ff19a7c -
diag: add key count to /health
b46e1b9 -
fix(nim): deep audit fixes — activate rotator >=1 key, catch transient errors per-key, verbose key-level logging
b732b30 -
fix(nim): uncap key rotator retries to cycle through all healthy keys in pool on 429
05ca8ed -
fix(nim): support multi-prefix key discovery (NVIDIA_API_KEY_X, NVIDIA_NIM_X) and seamless 429 failover
eb11d34
-
fix(nim): harden _NormalizedKimiStream against all tool_call object types
9ca8284 -
fix(nim): assign unique UUID tool call IDs per tool invocation in Kimi stream to prevent multi-turn ID collisions
af3b54d -
fix(nim): sanitize Kimi K3 tool IDs and suppress unsolicited thinking blocks on non-thinking requests
542f344 -
feat: upstream sync to 7fedb42 + Py3.12 compatibility and custom HF addons
3934d27
-
fix: force top_p=0.95 for Kimi K2/K3 models on NIM
6f345d2
-
fix: increase progress timeout to 900s, add SSE keepalives, and expose PROVIDER_PROGRESS_TIMEOUT
28035e3 -
feat: accept x-api-key and bearer token in require_proxy_auth
c7221ec -
fix: set requires-python >=3.12 for Python 3.12 Docker environment
d4d36ad -
fix: add Hugging Face Spaces Docker YAML frontmatter to README.md
726ed97 -
feat: sync upstream, add sol/terra/luna routing, set deepseek fallback, and fix Python 3.12 compatibility
abe1c43
-
fix(streams): use explicit class calls in ManagedStreamingResponse and provider cleanup
c4d5f12 -
fix(nim): align _create_stream signature and retry session handling with upstream OpenAIChatProvider
9cb1e44 -
fix(core): use object.__setattr__ in ExecutionFailure to resolve super() type error
3528434 -
chore: sync with upstream and apply custom HF proxy addons
261c237
-
fix(nim): dynamically clamp max_tokens to resolve context window exhaustion (400 Bad Request)
3f2f05a -
fix(nim): classify expected context length / max context 400/500 errors as context_window_exceeded for automatic client compaction
ea0ba58 -
fix(provider): remove invalid flush_task_arg_buffers call causing HTTP 500 error on streaming responses
58f7718 -
feat: sync upstream commits up to f17c92bc
- 51cc4be3: detect NIM 500 context overflow as context_window_exceeded (enables instant client compaction on NIM 500 max_tokens errors) - cc9fc53c & 58147b1e: add Kilo.ai provider with protocol-correct discovery and reasoning replay - 1622523b: add ChatGPT OAuth / OpenAI Codex provider - f17c92bc: add Azure OpenAI provider - Python 3.12 compatibility: parenthesized multi-exception tuples and module-level future annotations across all filesa1bab2d