Morph MORPH
SIGN IN SIGN UP
dhruvhead / vmlx UNCLAIMED
0 0 80 Python

COMMITS

main
August 22, 2026
J
feat(vlm): chunk the media-expanded prefill so a long VL chat stops hitting a wall
Jinho Jang committed
831a305
J
fix(cache): store the model's NATIVE cache representation, and stop imposing q4 on Nemotron Omni
Jinho Jang committed
276574c
J
fix(server): strip visual-grounding markup in the tool-suppression branch too
Jinho Jang committed
a94a6f4
J
fix(vlm): the image-prefill budget advises, it no longer refuses
Jinho Jang committed
841925d
J
chore(cache): make the SSM key debug line self-identifying
Jinho Jang committed
546a5a0
J
chore(cache): promote the SSM companion key/store hashes to INFO under VMLX_CACHE_HASH_DEBUG
Jinho Jang committed
362c80b
J
fix(mllm): bind the clean-store key for every request, not only the finishing one
Jinho Jang committed
04313d3
J
fix(mllm-cache): block-align the media clean boundary so KV and the SSM companion can actually meet
Jinho Jang committed
524d9c3
J
fix(cache): one shared prefix-hit helper, media-history turns reuse their text prefix, DSV4 block size reconciled where every launcher converges
Jinho Jang committed
0c89045
J
fix(mllm-cache): media turns can capture companions again, and every companion store now uses the key its reader asks with
Jinho Jang committed
dc56278
J
fix(cache): attempt the companion delta BEFORE the block table is trimmed
Jinho Jang committed
e89006f
J
feat(mtp): confidence gate that stops extending a draft chain the head is unsure about
Jinho Jang committed
0655d33
J
fix: never run an external drafter and the bundle's MTP heads together
Jinho Jang committed
d54add1
J
fix(cli): match the panel's stream-interval default instead of shipping the bad one
Jinho Jang committed
7f582f6
J
perf(cache): stop discarding a whole hybrid KV hit over a 31-token misalignment
Jinho Jang committed
214e3f6
J
perf(cache): stop re-running the whole prefix to attach an SSM companion
Jinho Jang committed
2327f06
J
fix(cache): say what the stored cache actually does, and bench it that way
Jinho Jang committed
6c7f50c
J
fix(cache): recover multiturn reuse on rotating-SWA instead of going cold
Jinho Jang committed
2c931e3
J
diag(cache): log which tier answered a block lookup, behind VMLX_CACHE_HASH_DEBUG
Jinho Jang committed
614f89d
J
fix(cache): preserve a deliberate "Unlimited", and stop a bf16 vision embedding crashing the cache key
Jinho Jang committed
e5b3319
August 21, 2026
J
fix(cache): stop shipping an unbounded SSD cache, and prepare the 1x bf16 block path
Jinho Jang committed
41035e7
J
feat(panel): say the in-memory cache default CHANGED to off, and that the SSD pool spans cache architectures
Jinho Jang committed
7a6b804
J
fix(cache): register the percent flag everywhere, tell users when their budget is too small
Jinho Jang committed
ae0565f
J
test(panel): assert the SSD budget reaches argv against the real launcher
Jinho Jang committed
133ac98
J
fix(cache): make the SSD budget real end-to-end and stop v17 corrupting migration fingerprints
Jinho Jang committed
9200e6e
J
feat(panel): tell every user what the in-memory cache costs, at the top of Session Settings
Jinho Jang committed
ae47fcf
J
feat(panel): warn low-memory Macs about the in-RAM cache, in every language
Jinho Jang committed
742cb67
J
chore(release): bump uv.lock to 1.6.35
Jinho Jang committed
2612824
J
chore(release): 1.6.35
Jinho Jang committed
69e7240
J
perf(cache): assemble the clean prompt-boundary cache instead of re-prefilling
Jinho Jang committed
f5b8228