Commit Graph
8 Commits
Author SHA1 Message Date
Luigi Colluto f02d79cb6b Reject oversized snapshots before allocation
Discard invalid distributed snapshot bodies before reserving token memory or entering the token read loop.

Closes #546
2026-08-04 13:23:59 +02:00
antirez 005afedcd9 Add GLM 5.2 inference and quality fixtures 2026-07-18 18:48:45 +02:00
Salvatore Sanfilippo 5cd0739c47 Fix distributed KV snapshot request IDs 2026-06-05 11:35:18 +02:00
antirez f7511c228b cuda: keep Flash paths building after PRO Q4 API changes 2026-06-02 09:07:28 +02:00
user 3c3ad4c647 Support distributed DeepSeek V4 PRO Q4 inference
Metal execution

Add the PRO Q4 routed-expert Metal path and distributed runtime changes needed to run the full model across two hosts.

Sliced GGUF loading

Allow distributed engines to bind only their local layer range, with token embeddings and the output head required only on the sides that need them.

Split artifacts

Document the two-file PRO Q4 setup and add download_model.sh targets for the coordinator half, worker half, and combined split download.

(cherry picked from commit a782cfba894c6a44af11e1b5fc69ccfc000ab39d)
2026-06-02 09:07:28 +02:00
user fec25dafdf build: fix current compiler warnings
(cherry picked from commit 0b3efaf86f61421330e90629508adbd6228b4a8b)
2026-05-29 18:50:54 +02:00
Salvatore Sanfilippo 4844d55108 Add distributed KV checkpoint support 2026-05-29 17:53:57 +02:00
Salvatore Sanfilippo abdc807ced Add distributed inference
Add coordinator/worker distributed layer execution, pipelined prefill, worker routing, telemetry, activation transport width, and KV mismatch recovery for DeepSeek Flash/Pro.
2026-05-29 12:12:10 +02:00