Morph MORPH
SIGN IN SIGN UP
dhruvhead / speculative-decoding UNCLAIMED
0 0 31 Swift

COMMITS

master
December 8, 2025
L
Update benchmarks with M4 Pro results and add Mamba vocab mismatch note
lulzx committed
3aa02aa
L
Revert to Qwen2.5 models (Qwen3 not yet supported in MLX-Swift)
lulzx committed
61d17f6
L
Update benchmark results with M4 Pro numbers
lulzx committed
2a21776
L
Fix Swift 6 Sendable warnings
lulzx committed
d981673
L
Update CLI default models to Qwen3
lulzx committed
e1fe2f3
L
Update recommended model pairs to Qwen3, gemma-3, SmolLM3
lulzx committed
b18bb50
L
Update README to use Qwen3, gemma-3, and SmolLM3 models
lulzx committed
776989e
L
Update README with Mamba benchmark results
lulzx committed
e41dc22
L
Add benchmark command for Mamba vs Transformer comparison
lulzx committed
efda5b6
L
Add unit tests for Mamba components
lulzx committed
bc0e80c
L
Add Mamba drafter and speculative generator
lulzx committed
659f729
L
Add HuggingFace weight loader for Mamba models
lulzx committed
9829e9e
L
Add Mamba language model architecture
lulzx committed
4f878c0
L
Add Mamba selective SSM block
lulzx committed
a0f786f
L
Add constant-size cache for Mamba inference
lulzx committed
e3c5a9b
L
Add Mamba configuration for SSM models
lulzx committed
ebeb27c
November 28, 2025
L
Add README and update gitignore
lulzx committed
567f3b4
L
Add unit tests for verification sampler
lulzx committed
5e75afb
L
Add CLI example tool
lulzx committed
82f2072
L
Add SpeculativeDecoding public API
lulzx committed
795f238
L
Add SpeculativeGenerator with main draft-verify loop
lulzx committed
3965dcf
L
Add VerificationSampler with rejection sampling algorithm
lulzx committed
69e847c
L
Add SpeculativeCache for dual KV-cache management
lulzx committed
84b60a6
L
Add DraftTargetPair for model loading and management
lulzx committed
66cf6e1
L
Add SpeculativeParameters for generation configuration
lulzx committed
cac9963
L
Add Package.swift with MLX-Swift dependencies
lulzx committed
07bfb3a
L
Initial commit
Lulzx committed
c125b80