v0.2.0

What's Changed

• fix llama3.1 by @wejoncy in #132
• support transformers-lib loading by @wejoncy in #134
• bump to 0.2.0 by @wejoncy in #135

Full Changelog: v0.1.9.1...v0.2.0

v0.1.9.1

What's Changed

• add assert message && ci upgrade torch 2.2.2 by @wejoncy in #124
• Update README.md by @wejoncy in #125
• fix version match erros by @wejoncy in #128
• add macro GENERAL_TORCH to get rid of OptionalCUDAGuard by @wejoncy in #129
• quick fix by @wejoncy in #130
• v0.1.9.1 by @wejoncy in #131

Full Changelog: v0.1.9...v0.1.9.1

v0.1.9

What's Changed

• Bump to 0.1.8 by @wejoncy in #109
• new autogptq config format && parallel load by @wejoncy in #110
• bugfix by @wejoncy in #111
• fix issue by @wejoncy in #113
• Fix 112 by @wejoncy in #114
• Fix typos by @emphasis10 in #115
• minor fix, attn_implementation by @wejoncy in #120
• Bump to 0.1.9 by @wejoncy in #121
• -allow-unsupported-compiler by @wejoncy in #122

New Contributors

• @emphasis10 made their first contribution in #115

Full Changelog: v0.1.8...v0.1.9

v0.1.8

What's Changed

• Update README.md by @wejoncy in #102
• buf fix. by @wejoncy in #103
• Onnx fix qzeros odd-shape by @wejoncy in #104
• Refactor by @wejoncy in #105
• support MARLIN pack_mode by @wejoncy in #106
• support awq sym by @wejoncy in #107
• Refactor by @wejoncy in #108

Full Changelog: v0.1.7.1...v0.1.8

v0.1.7.1

What's Changed

• fix "disable win in release by @wejoncy in #98
• minor fix and dataset speed by @wejoncy in #99
• minor fix by @wejoncy in #100
• patch release v0.1.7.1 by @wejoncy in #101

Full Changelog: v0.1.7...v0.1.7.1

v0.1.7

What's Changed

• ort ops support in main branch with act_order by @wejoncy in #92
• support export hqq to onnx by @wejoncy in #93
• Bump to 0.1.7 by @wejoncy in #94
• improve .cpu() with non_blocking by @wejoncy in #95
• disable win in release by @wejoncy in #96
• refactor args by @wejoncy in #97

Full Changelog: v0.1.6...v0.1.7

v0.1.6

What's Changed

• illegal memory access by @wejoncy in #71
• Format by @wejoncy in #72
• Hqq support by @wejoncy in #73
• [feature]fast build support [a few seconds to build and install qllm ] by @wejoncy in #74
• minor fix for detecting ort_ops and torch.compile by @wejoncy in #75
• static_groups by @wejoncy in #76
• minor Fix (cudaguard) by @wejoncy in #77
• fix cudaguard by @wejoncy in #78
• ruff Format by @wejoncy in #79
• move ops into qllm by @wejoncy in #80
• fix memory layout in QuantizeLinear by @yufenglee in #82
• add continuous check for ort kernel by @wejoncy in #84
• Ort fix by @wejoncy in #85
• more general Ort ops export by @wejoncy in #86
• 0.1.6.dev by @wejoncy in #87
• 0.1.6 by @wejoncy in #88
• speed up ort node packing by @wejoncy in #89
• fix attn_implementation by @wejoncy in #90

New Contributors

• @yufenglee made their first contribution in #82

Full Changelog: v0.1.5...v0.1.6

v0.1.5

What's Changed

• works on windows, set dtype is importang by @wejoncy in #54
• use_heuristic=false default for models having hard to predict unquantized layers like mixtral-8x7b by @wejoncy in #55
• add mixtral in readme example by @wejoncy in #56
• bugfix when export 16bit model by @wejoncy in #57
• Fix build err, uint32_t is not defined. <stdint.h> by @wejoncy in #58
• dp kernel support g_idx by @wejoncy in #59
• [important] packing improve, faster by @wejoncy in #60
• [improve packing]fix for awq unpack by @wejoncy in #61
• 3bit support with g_idx in dq_kernel by @wejoncy in #63
• 3bit fix by @wejoncy in #64
• 0.1.5.dev by @wejoncy in #65
• onnx support Act_order && some onnx fix by @wejoncy in #66
• Support gemv with g_idx and some fix in exporter/dataloader by @wejoncy in #67
• support mixtral in gptq/awq by @wejoncy in #68
• minor fix for act_order detect by @wejoncy in #70
• Bump version to 0.1.5 by @wejoncy in #69

Full Changelog: v0.1....v0.1.5

v0.1.4

What's Changed

• suport Phi, detect multi blocks by @wejoncy in #43
• quick fix by @wejoncy in #44
• add colab example && turing support for awq && remove dependency of xbitops by @wejoncy in #46
• quick fix for meta device by @wejoncy in #47
• add trust code by @wejoncy in #48
• fix trust_code by @wejoncy in #49
• quick fix for turing awq 75 by @wejoncy in #50
• fix low_cpu_mem_usage by @wejoncy in #51
• fix model dtype ,default half by @wejoncy in #52

Full Changelog: v0.1.3...v0.1.4

v0.1.3

What's Changed

• Bump version to 0.1.3 by @wejoncy in #29
• pipeline by @wejoncy in #33
• minor fix win/special by @wejoncy in #34
• Fix pack_mode issue and add Proxy for transformers. by @wejoncy in #36
• work around autoqptq/vLLM by @wejoncy in #37
• update readme and fix some pack_mode conversion bugs by @wejoncy in #38
• minor fix and rename quat_linear folders by @wejoncy in #39
• [fix] Weight pack && tokenizer && more awq models by @wejoncy in #40
• Readme by @wejoncy in #41
• ready for pypi package by @wejoncy in #42

Full Changelog: https://github.com/wejoncy/QLLM/commits/v0.1.3