• Joined on 2026-02-17
grenade pushed to feat/computed-context-limits at helexa/helexa 2026-06-17 11:02:07 +00:00
464b6b0db9 feat(neuron): self-measured prefill tok/s EMA on streaming paths (#67 phase 3)
grenade pushed to main at gongfoo/gongfoo 2026-06-17 10:48:31 +00:00
546c9b8705 fix(runner): raise cuda-13.0 memory to 32 GiB to clear flash-attn OOM
grenade pushed to feat/computed-context-limits at helexa/helexa 2026-06-17 10:18:32 +00:00
f2e05d96ec feat(neuron): capture ContextProfile at load + per-rank VRAM fan-out (#67 phase 2)
grenade created branch feat/computed-context-limits in helexa/helexa 2026-06-17 10:00:56 +00:00
grenade pushed to feat/computed-context-limits at helexa/helexa 2026-06-17 10:00:56 +00:00
4f05a87449 feat(neuron): self-derived context-limit core — physics + policy (#67 phase 1)
grenade opened issue helexa/helexa#68 2026-06-17 09:22:30 +00:00
investigate: the cost/pricing schema on /v1/models — value source + a predictable, client-consumable shape
grenade deleted branch feat/usage-reasoning-tokens from helexa/helexa 2026-06-17 09:11:50 +00:00
grenade closed issue helexa/helexa#64 2026-06-17 09:11:09 +00:00
neuron: emit reasoning_tokens / cached_tokens usage details
grenade pushed to main at helexa/helexa 2026-06-17 09:11:09 +00:00
2f67d17ec7 feat(neuron): emit reasoning_tokens usage details on streaming
grenade created branch feat/usage-reasoning-tokens in helexa/helexa 2026-06-17 09:04:16 +00:00
grenade pushed to feat/usage-reasoning-tokens at helexa/helexa 2026-06-17 09:04:16 +00:00
2f67d17ec7 feat(neuron): emit reasoning_tokens usage details on streaming
grenade pushed to main at lair/mistralrs-package 2026-06-17 08:56:52 +00:00
aae3cf3141 chore(build): CARGO_TERM_COLOR=always
grenade pushed to main at lair/mistralrs-package 2026-06-17 08:46:37 +00:00
d322943fa9 fix(build): pass -Wno-template-body to nvcc host compiler for gcc 15
grenade commented on issue helexa/helexa#64 2026-06-17 08:16:06 +00:00
neuron: emit reasoning_tokens / cached_tokens usage details

Scoped implementation plan (streaming-only reasoning_tokens)

File:line map so this is pick-up-able. The hard part — reasoning-span detection — already exists (in_reasoning is tracked…

grenade pushed to main at helexa/helexa 2026-06-17 07:04:32 +00:00
11b2e6f78c fix(cortex): default models_config to the packaged absolute path
grenade commented on issue helexa/helexa#67 2026-06-17 06:57:21 +00:00
neuron: self-derive and advertise optimal limit{context,input,output} from model config + live VRAM; catalogue limit becomes an override

Acceptance addition: removal of the temporary operator override is the validation

To unblock the opencode helexa-provider implementation now (before this lands), a deliberate, temporary

grenade opened issue helexa/helexa#67 2026-06-17 06:37:27 +00:00
neuron: self-derive and advertise optimal limit{context,input,output} from model config + live VRAM; catalogue limit becomes an override
grenade pushed to main at helexa/helexa 2026-06-17 06:34:01 +00:00
8a636c687f feat(cortex): per-model limit + cost on /v1/models; remove max_model_len
grenade closed issue helexa/helexa#62 2026-06-17 06:34:00 +00:00
neuron: enrich /models with per-model limit + cost (opencode catalog discovery)
grenade closed 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)