Block a user
grenade
pushed to fix/followers-stream-from-the-transfer at grenade/rustingface
2026-09-02 11:34:22 +00:00
grenade
created branch fix/followers-stream-from-the-transfer in grenade/rustingface
2026-09-02 11:34:21 +00:00
A cold prefix range delivers its bytes at once, then stalls on the holdback
grenade
created branch fix/bounded-cold-range-does-not-stall in grenade/rustingface
2026-09-02 11:24:55 +00:00
grenade
pushed to fix/bounded-cold-range-does-not-stall at grenade/rustingface
2026-09-02 11:24:55 +00:00
A deploy during a long download kills the service with SIGABRT
grenade
pushed to fix/drain-does-not-wait-on-blob-streams at grenade/rustingface
2026-09-02 11:16:56 +00:00
grenade
created branch fix/drain-does-not-wait-on-blob-streams in grenade/rustingface
2026-09-02 11:16:55 +00:00
A cold prefix range delivers its bytes at once, then stalls on the holdback
A deploy during a long download kills the service with SIGABRT
A request that joins an in-flight fetch waits in silence for the whole transfer
Measured the pinned client before building on the suggestion above, and it does not hold up. Correcting it.
huggingface_hub 0.36 routes file requests through
_request_wrapper -> http_backoff…
A request that joins an in-flight fetch waits in silence for the whole transfer
A total-duration request_timeout makes large-blob fetches fail deterministically in proxy mode
qwen4_exp F: use the shipped MTP head as a draft model
Settled against vLLM — and the spec had one of the norms wrong
Cloned vllm-project/vllm (at 76ba321). The head is implemented at
vllm/models/qwen4_exp/nvidia/mtp.py. vLLM calls the…