grenade

grenade commented on issue helexa/helexa#66

2026-06-17 06:09:05 +00:00

cortex: re-discover a neuron's max_prompt_tokens after it restarts (stale max_model_len on /v1/models)

Closing as moot — the premise doesn't hold.

#66 assumed cortex caching a stale max_model_len after a neuron restart had client impact. But max_model_len has no known consumer: opencode's…

grenade opened issue helexa/helexa#66

2026-06-16 15:51:17 +00:00

cortex: re-discover a neuron's max_prompt_tokens after it restarts (stale max_model_len on /v1/models)

grenade pushed to main at helexa/helexa

2026-06-16 15:48:23 +00:00

6088830e7d feat(deploy): manage NEURON_MAX_PROMPT_TOKENS per host via model.conf drop-in

grenade opened issue helexa/helexa#65

2026-06-16 15:18:30 +00:00

neuron: length-aware VRAM pre-flight for text prefill (mirror vision guard) — prevents OOM-poison at high NEURON_MAX_PROMPT_TOKENS

grenade closed issue helexa/helexa#63

2026-06-16 14:51:17 +00:00

cortex/neuron: unify all rejection responses on the OpenAI error envelope + standard codes

grenade pushed to main at helexa/helexa

2026-06-16 14:51:17 +00:00

04f798ec23 feat(cortex-gateway): enhance error responses with structured data

grenade closed issue helexa/helexa#48

2026-06-16 11:23:51 +00:00

neuron: emit usage (prompt/completion tokens) on SSE streams

grenade opened issue helexa/helexa#64

2026-06-16 11:23:40 +00:00

neuron: emit reasoning_tokens / cached_tokens usage details

grenade pushed to main at helexa/helexa

2026-06-16 11:15:36 +00:00

6f3e9276cd docs: add AGENTS.md with project architecture, build commands, and conventions

grenade opened issue helexa/helexa#63

2026-06-16 09:30:12 +00:00

cortex/neuron: unify all rejection responses on the OpenAI error envelope + standard codes

grenade opened issue helexa/helexa#62

2026-06-16 09:00:29 +00:00

neuron: enrich /models with per-model limit + cost (opencode catalog discovery)

grenade opened issue helexa/helexa#61

2026-06-16 09:00:04 +00:00

Epic: first-class opencode integration (dynamic helexa provider)

grenade pushed to main at lair/mistralrs-package

2026-06-16 08:05:38 +00:00

ca69a6fdfb chore(build-prerelease): reduce build jobs and nvcc threads

grenade commented on issue helexa/helexa#60

2026-06-15 17:50:14 +00:00

neuron: emit OpenAI-standard error envelope for all InferenceError variants (fix context_length_exceeded detection)

✅ CI run #506 green on 8f9e956 — Lint (fmt + clippy), Test, and all three neuron builds (ada/ampere/blackwell) + RPM packages passed.…

grenade closed issue helexa/helexa#60

2026-06-15 17:50:08 +00:00

neuron: emit OpenAI-standard error envelope for all InferenceError variants (fix context_length_exceeded detection)

grenade commented on issue helexa/helexa#60

2026-06-15 17:42:46 +00:00

neuron: emit OpenAI-standard error envelope for all InferenceError variants (fix context_length_exceeded detection)

Implemented in 8f9e956.

neuron now emits the OpenAI-standard nested error envelope for every InferenceError:

{ "error": { "message": "...", "type": "...", "code": "...", "param":…

grenade pushed to main at helexa/helexa

2026-06-15 17:42:21 +00:00

8f9e956d17 fix(neuron): emit OpenAI-standard nested error envelopes (#60)

grenade opened issue helexa/helexa#60

2026-06-15 17:27:41 +00:00

neuron: emit OpenAI-standard error envelope for all InferenceError variants (fix context_length_exceeded detection)

grenade pushed to main at helexa/helexa

2026-06-15 16:44:03 +00:00

cb758d4706 feat(neuron): emit usage on the streaming path so clients can track context

grenade pushed to main at helexa/helexa

2026-06-15 16:11:17 +00:00

a2d2dbd006 feat: advertise max_model_len on /v1/models so clients can compact