Insights DB

avlp12 / GLM-5.3-Flash-Alis-MLX-4bit

Last scanned: Sep 1, 2026 at 12:07 AM UTC

Report an issue

Versions

Version	Commit message	Commit author	Last scanned
97b6cf	Card update 2026-09-07: 5-bit-KDA build served locally (repo files unchanged), new speculative headline, thinking budget with speculation, disk prefix vault; earlier sections unchanged	avlp12	Never scanned
e9f560	Card: date-label the historical sections (as of 2026-09-05) so they are not read as current; numbers unchanged	avlp12
2691c6	Card update 2026-09-06: new measurements section first (decode/speculative/sampled/structured/prefix-cache/prefill/pipeline-parallel/ceilings), corrections, negative results; earlier content unchanged	avlp12
3e2c84	card: condition-labelled decode/speculative rows; withdraw identical-row B=8 aggregate (paired +49% kept), unprovenanced 1.64x TP-spec headline, and the context-flat claim; add speculative-prefill limit (32k) and non-identity caveats; label Metal buffer settings; #2105 status	avlp12
e4c9f9	prefill@131k 295 tok/s (1.33-1.35x paired x2 loads); spec 50.8 code with adaptive width + refit constants	avlp12
445e4e	TP+spec re-validated: 49.8 tok/s (1.44x on top of TP, prose 1.32x); 60.8 retired	avlp12
bbcb7e	speed section speculative bullet corrected to natural-prompt figures	avlp12
4164b2	speculative rows corrected: 45.9 (1.51x) natural code replaces PAD-flattered 54.7; TP+spec marked re-validation pending	avlp12
3b2a2b	TP setup: note FAST_SYNCH not-guaranteed status per mlx#4438 maintainer reply; recommend per-fleet soak validation	avlp12
b35653	TP cells closed: B=8 162 and prefill 631/910 re-validated post-fix; B=1 both-accountings noted	avlp12
c2e7c6	TP numbers corrected after indexer-reduction fix: B=1 36.2 (+12%), TP+spec 60.8 (1.88x); B=8/prefill re-validation pending	avlp12
ebbc93	remove reference to private conversion-pipeline repo	avlp12
fb54b8	TP=2 two-box serving: +27% B=1 / +48% B=8 / -48% memory per box; spec composes to 1.96x	avlp12
7005b9	card: link the fused-KDA upstream PR (mlx-vlm#2105)	avlp12
729092	card: fused-KDA kernels now public (avlp12/mlx-vlm glm5-serve-unified) with one-install reproduction	avlp12
217629	card: at-a-glance table+chart, gate 32768 numbers, reproducible serving guide	avlp12
ec51bc	card: final campaign numbers (fused-KDA kernel 34.8 tok/s, #2087 long-context prefill +50%)	avlp12
e6ae4b	Squash history (withdrawn-build blobs GC)	avlp12

‌
‌
‌
‌
‌
‌
‌
‌
‌
‌

Prisma AIRS detects threats in model files

With Prisma AIRS you can scan models for threats before ML developers download them for use, and apply policies based on your risk tolerance.

Learn more

Found a new threat?

Earn bounties and get recognition for your discoveries by submitting them through Huntr.

Report your finding

Version	Commit message	Commit author	Last scanned
97b6cf	Card update 2026-09-07: 5-bit-KDA build served locally (repo files unchanged), new speculative headline, thinking budget with speculation, disk prefix vault; earlier sections unchanged	avlp12	Never scanned
e9f560	Card: date-label the historical sections (as of 2026-09-05) so they are not read as current; numbers unchanged	avlp12
2691c6	Card update 2026-09-06: new measurements section first (decode/speculative/sampled/structured/prefix-cache/prefill/pipeline-parallel/ceilings), corrections, negative results; earlier content unchanged	avlp12
3e2c84	card: condition-labelled decode/speculative rows; withdraw identical-row B=8 aggregate (paired +49% kept), unprovenanced 1.64x TP-spec headline, and the context-flat claim; add speculative-prefill limit (32k) and non-identity caveats; label Metal buffer settings; #2105 status	avlp12
e4c9f9	prefill@131k 295 tok/s (1.33-1.35x paired x2 loads); spec 50.8 code with adaptive width + refit constants	avlp12
445e4e	TP+spec re-validated: 49.8 tok/s (1.44x on top of TP, prose 1.32x); 60.8 retired	avlp12
bbcb7e	speed section speculative bullet corrected to natural-prompt figures	avlp12
4164b2	speculative rows corrected: 45.9 (1.51x) natural code replaces PAD-flattered 54.7; TP+spec marked re-validation pending	avlp12
3b2a2b	TP setup: note FAST_SYNCH not-guaranteed status per mlx#4438 maintainer reply; recommend per-fleet soak validation	avlp12
b35653	TP cells closed: B=8 162 and prefill 631/910 re-validated post-fix; B=1 both-accountings noted	avlp12
c2e7c6	TP numbers corrected after indexer-reduction fix: B=1 36.2 (+12%), TP+spec 60.8 (1.88x); B=8/prefill re-validation pending	avlp12
ebbc93	remove reference to private conversion-pipeline repo	avlp12
fb54b8	TP=2 two-box serving: +27% B=1 / +48% B=8 / -48% memory per box; spec composes to 1.96x	avlp12
7005b9	card: link the fused-KDA upstream PR (mlx-vlm#2105)	avlp12
729092	card: fused-KDA kernels now public (avlp12/mlx-vlm glm5-serve-unified) with one-install reproduction	avlp12
217629	card: at-a-glance table+chart, gate 32768 numbers, reproducible serving guide	avlp12
ec51bc	card: final campaign numbers (fused-KDA kernel 34.8 tok/s, #2087 long-context prefill +50%)	avlp12
e6ae4b	Squash history (withdrawn-build blobs GC)	avlp12

Version

Commit message

Commit author

Last scanned

97b6cf

Card update 2026-09-07: 5-bit-KDA build served locally (repo files unchanged), new speculative headline, thinking budget with speculation, disk prefix vault; earlier sections unchanged

avlp12

Never scanned

e9f560

Card: date-label the historical sections (as of 2026-09-05) so they are not read as current; numbers unchanged

avlp12

2691c6

Card update 2026-09-06: new measurements section first (decode/speculative/sampled/structured/prefix-cache/prefill/pipeline-parallel/ceilings), corrections, negative results; earlier content unchanged

avlp12

3e2c84

card: condition-labelled decode/speculative rows; withdraw identical-row B=8 aggregate (paired +49% kept), unprovenanced 1.64x TP-spec headline, and the context-flat claim; add speculative-prefill limit (32k) and non-identity caveats; label Metal buffer settings; #2105 status

avlp12

e4c9f9

prefill@131k 295 tok/s (1.33-1.35x paired x2 loads); spec 50.8 code with adaptive width + refit constants

avlp12

445e4e

TP+spec re-validated: 49.8 tok/s (1.44x on top of TP, prose 1.32x); 60.8 retired

avlp12

bbcb7e

speed section speculative bullet corrected to natural-prompt figures

avlp12

4164b2

speculative rows corrected: 45.9 (1.51x) natural code replaces PAD-flattered 54.7; TP+spec marked re-validation pending

avlp12

3b2a2b

TP setup: note FAST_SYNCH not-guaranteed status per mlx#4438 maintainer reply; recommend per-fleet soak validation

avlp12

b35653

TP cells closed: B=8 162 and prefill 631/910 re-validated post-fix; B=1 both-accountings noted

avlp12

c2e7c6

TP numbers corrected after indexer-reduction fix: B=1 36.2 (+12%), TP+spec 60.8 (1.88x); B=8/prefill re-validation pending

avlp12

ebbc93

remove reference to private conversion-pipeline repo

avlp12

fb54b8

TP=2 two-box serving: +27% B=1 / +48% B=8 / -48% memory per box; spec composes to 1.96x

avlp12

7005b9

card: link the fused-KDA upstream PR (mlx-vlm#2105)

avlp12

729092

card: fused-KDA kernels now public (avlp12/mlx-vlm glm5-serve-unified) with one-install reproduction

avlp12

217629

card: at-a-glance table+chart, gate 32768 numbers, reproducible serving guide

avlp12

ec51bc

card: final campaign numbers (fused-KDA kernel 34.8 tok/s, #2087 long-context prefill +50%)

avlp12

e6ae4b

Squash history (withdrawn-build blobs GC)

avlp12

avlp12 / GLM-5.3-Flash-Alis-MLX-4bit

Versions

Prisma AIRS detects threats in model files

Found a new threat?

Products and Services

Company

Popular Links