Repository navigation

Quant - #8

Merged

ravi9 merged 30 commits into

dev_backend_openvinofrom

quant

Sep 25, 2025

Merged

Quant#8
ravi9 merged 30 commits into
dev_backend_openvinofrom
quant

Commits on Sep 8, 2025
- Update supports_buft and supports_op for quantized models
  wine99
  committed
  08de982
- Add quant weight conversion functions from genai gguf reader
  wine99
  committed
  7cabe37
- Quant models run with accuracy issue
  wine99
  committed
  976c85b
- Fix accuracy: disable cpu_repack
  wine99
  committed
  1f8e007
- Fix CI; Disable test-backend-ops
  wine99
  committed
  c77cefe
- Fix Q4_1
  wine99
  committed
  3023981
- Fix test-thread-safety
  wine99
  committed
  7b8189a
- Fix test-backend-ops: Treat quantized tensors as weights
  wine99
  committed
  0376a7a
- Add NPU Q4_0 support
  wine99
  committed
  e1a5f7e
- NPU perf: eliminate zp
  wine99
  committed
  489d453
- NPU perf: Faster compilation
  wine99
  committed
  c319ce5
- Dequantize q4_1 q4_k q6_k for NPU
  wine99
  committed
  5b6418d
- Add custom quant type: q8_1_c, q4_0_128
  wine99
  committed
  623f863
- Set m_is_static=false as default in decoder
  wine99
  committed
  ef4de4d
- Simpilfy translation of get_rows
  wine99
  committed
  1b8323f
- Fix after rebasing
  wine99
  committed
  6da5a22
Commits on Sep 10, 2025
- Improve debug util; Eliminate nop ReshapeReshape
  wine99
  committed
  e37cdd4
- STYLE: make get_types_to_requant a function
  wine99
  committed
  b453d68
Commits on Sep 11, 2025
- Support BF16 model
  wine99
  committed
  4ed2510
Commits on Sep 12, 2025
- Fix NPU compile
  wine99
  committed
  3eeb567
- WA for npu 1st token acc issue
  wine99
  committed
  3346a33
- Apply EliminateZP only for npu
  wine99
  committed
  fa237a1
Commits on Sep 15, 2025
- Add GeGLU
  wine99
  committed
  9a85e53
- Fix Hunyuan
  wine99
  committed
  3d31fa6
Commits on Sep 16, 2025
- Support iSWA
  wine99
  committed
  500aead
Commits on Sep 17, 2025
- Fix NPU accuracy
  wine99
  committed
  a079242
- Fix ROPE accuracy when freq_scale != 1
  wine99
  committed
  b6c84af
- Minor: not add attention_size_swa for non-swa model
  wine99
  committed
  b3eb6fb
Commits on Sep 19, 2025
- Minor refactor
  wine99
  committed
  bb33530
Commits on Sep 23, 2025
- Add Q5_K to support phi-3-q4_k_m
  wine99
  committed
  812590b

Commits

Commits on Sep 8, 2025

Commits on Sep 10, 2025

Commits on Sep 11, 2025

Commits on Sep 12, 2025

Commits on Sep 15, 2025

Commits on Sep 16, 2025

Commits on Sep 17, 2025

Commits on Sep 19, 2025

Commits on Sep 23, 2025