Commits

Branch selector

 master

User selector

ikawrakow

Datepicker

All time

Commit History

Commits on Mar 27, 2024

• Make IQ1_M work for QK_K = 64 (ggerganov#6327)

  Show description for cbc8343
  

  ikawrakow

  and

  Kawrakow

  authored
  cbc8343

  Copy full SHA for cbc8343

Commits on Mar 26, 2024

• IQ1_M: 1.75 bpw quantization (ggerganov#6302)

  Show description for 55c1b2a
  

  ikawrakow

  and

  Kawrakow

  authored
  55c1b2a

  Copy full SHA for 55c1b2a

• quantize : be able to override metadata by key (ggerganov#6321)

  Show description for d25b1c3
  

  3 peopleauthored
  d25b1c3

  Copy full SHA for d25b1c3

Commits on Mar 25, 2024

• tests : include IQ2_XXS and IQ2_XS in test-quantize-fns (ggerganov#6303)

  Show description for 1f2fd4e
  

  ikawrakow

  and

  Kawrakow

  authored
  1f2fd4e

  Copy full SHA for 1f2fd4e

Commits on Mar 22, 2024

• quantize: options for output and token embedding tensors qtype (ggerganov#6239)

  Show description for 1d0331c
  

  ikawrakow

  and

  Kawrakow

  authored
  1d0331c

  Copy full SHA for 1d0331c

Commits on Mar 21, 2024

• ggml : same IQ4_NL quantization for CPU/CUDA/Metal (ggerganov#6196)

  Show description for cfd3be7
  

  ikawrakow

  and

  Kawrakow

  authored
  cfd3be7

  Copy full SHA for cfd3be7

• Add ability to use Q5_0, Q5_1, and IQ4_NL for quantized K cache (ggerganov#6183)

  Show description for 76aa30a
  

  ikawrakow

  and

  Kawrakow

  authored
  76aa30a

  Copy full SHA for 76aa30a

Commits on Mar 11, 2024

• 1.5 bit: we can do even better (ggerganov#5999)

  Show description for 44ca159
  

  ikawrakow

  and

  Kawrakow

  authored
  44ca159

  Copy full SHA for 44ca159

• Better 1.5 bit quantization (ggerganov#5971)

  Show description for be858f6
  

  ikawrakow

  and

  Kawrakow

  authored
  be858f6

  Copy full SHA for be858f6

Commits on Mar 2, 2024

• ggml : IQ3_S improvements (ggerganov#5829)

  Show description for bbde6eb
  

  ikawrakow

  and

  Kawrakow

  authored
  bbde6eb

  Copy full SHA for bbde6eb

Commits on Feb 28, 2024

• ggml : make i-quants work with super-blocks of 64 (CPU,Metal) (ggerganov#5760)

  Show description for 7c4263d
  

  ikawrakow

  and

  Kawrakow

  authored
  7c4263d

  Copy full SHA for 7c4263d

Commits on Feb 27, 2024

• Attempt to fix android build (ggerganov#5752)

  Show description for cb49e0f
  

  ikawrakow

  and

  Kawrakow

  authored
  cb49e0f

  Copy full SHA for cb49e0f

• IQ4_XS: a 4.25 bpw quantization (ggerganov#5747)

  Show description for 0becb22
  

  ikawrakow

  and

  Kawrakow

  authored
  0becb22

  Copy full SHA for 0becb22

Commits on Feb 26, 2024

• Adding IQ2_S and IQ2_M to complete coverage of the 2-3 bit quantization range (ggerganov#5721)

  Show description for a33e6a0
  

  3 peopleauthored
  a33e6a0

  Copy full SHA for a33e6a0

Commits on Feb 24, 2024

• IQ3_S: a much better alternative to Q3_K (ggerganov#5676)

  Show description for 4c4cb30
  

  ikawrakow

  and

  Kawrakow

  authored
  4c4cb30

  Copy full SHA for 4c4cb30

Commits on Feb 21, 2024

• IQ4_NL: 4-bit non-linear quants with blocks of 32 (ggerganov#5590)

  Show description for a14679c
  

  ikawrakow

  and

  Kawrakow

  authored
  a14679c

  Copy full SHA for a14679c

Commits on Feb 18, 2024

• 1.5 bit quantization (ggerganov#5453)

  Show description for bd2d4e3
  

  ikawrakow

  and

  Kawrakow

  authored
  bd2d4e3

  Copy full SHA for bd2d4e3

Commits on Feb 13, 2024

• ggml-quants : fix compiler warnings (shadow variable) (ggerganov#5472)

  Show description for 895407f
  

  ikawrakow

  and

  Kawrakow

  authored
  895407f

  Copy full SHA for 895407f

Commits on Feb 6, 2024

• Update README.md (ggerganov#5366)

  Show description for b08f22c
  ikawrakow

  authored
  b08f22c

  Copy full SHA for b08f22c

• Slight quantization improvement for Q4_K and Q5_K (ggerganov#5361)

  Show description for f57fadc
  

  ikawrakow

  and

  Kawrakow

  authored
  f57fadc

  Copy full SHA for f57fadc

Commits on Feb 5, 2024

• ggml : make use of ggml-quants.h possible in C++ code (ggerganov#5338)

  Show description for c6b3955
  

  ikawrakow

  and

  Kawrakow

  authored
  c6b3955

  Copy full SHA for c6b3955

• iq3_xxs: quards for the no-imatrix situation (ggerganov#5334)

  Show description for 89503dc
  

  ikawrakow

  and

  Kawrakow

  authored
  89503dc

  Copy full SHA for 89503dc

• iq2_xxs: tune quantization (ggerganov#5320)

  Show description for 6fdfa2e
  

  ikawrakow

  and

  Kawrakow

  authored
  6fdfa2e

  Copy full SHA for 6fdfa2e

Commits on Feb 4, 2024

• Adding some imatrix tools (ggerganov#5302)

  Show description for 5ed26e1
  

  ikawrakow

  and

  Kawrakow

  authored
  5ed26e1

  Copy full SHA for 5ed26e1

Commits on Jan 30, 2024

• ggml : fix IQ3_XXS on Metal (ggerganov#5219)

  Show description for fea4fd4
  

  ikawrakow

  and

  Kawrakow

  authored
  fea4fd4

  Copy full SHA for fea4fd4

• Faster AVX2 dot product for IQ2_XS (ggerganov#5187)

  Show description for 8e14e3d
  

  3 peopleauthored
  8e14e3d

  Copy full SHA for 8e14e3d

• SOTA 3-bit quants (ggerganov#5196)

  Show description for f4d7e54
  

  ikawrakow

  and

  Kawrakow

  authored
  f4d7e54

  Copy full SHA for f4d7e54

Commits on Jan 26, 2024

• Another bucket sort (ggerganov#5109)

  Show description for 1182cf4
  

  ikawrakow

  and

  Kawrakow

  authored
  1182cf4

  Copy full SHA for 1182cf4

Commits on Jan 25, 2024

• Fix Q3_K_XS for MoE models (ggerganov#5113)

  Show description for faa3526
  

  ikawrakow

  and

  Kawrakow

  authored
  faa3526

  Copy full SHA for faa3526

Commits on Jan 23, 2024

• Additional KL-divergence statistics (ggerganov#5081)

  Show description for 44879ee
  

  ikawrakow

  and

  Kawrakow

  authored
  44879ee

  Copy full SHA for 44879ee

Commits on Jan 22, 2024

• KL-divergence (ggerganov#5076)

  Show description for 6f9939d
  

  ikawrakow

  and

  Kawrakow

  authored
  6f9939d

  Copy full SHA for 6f9939d

• imatrix : keep intermediate imatrix results (ggerganov#5077)

  Show description for 15bceec
  

  ikawrakow

  and

  Kawrakow

  authored
  15bceec

  Copy full SHA for 15bceec

• llama : add Q3_K_XS (ggerganov#5060)

  Show description for 66d575c
  

  ikawrakow

  and

  Kawrakow

  authored
  66d575c

  Copy full SHA for 66d575c

Commits on Jan 21, 2024

• Add ability to evauate multiple choice tasks (ggerganov#5047)

  Show description for 7dcbe39
  

  ikawrakow

  and

  Kawrakow

  authored
  7dcbe39

  Copy full SHA for 7dcbe39

• Slightly faster imatrix (ggerganov#5050)

  Show description for 726c0fa
  

  ikawrakow

  and

  Kawrakow

  authored
  726c0fa

  Copy full SHA for 726c0fa

Pagination

Previous

Next