* add export

* cleanup

* new nn.linear

* export to mlx

* move dequantize to troch linear

* cleanup

* desc_act=True is not supported

* fix lm_head

* should allow  gptq_v2

* add TestExport

* save tokenizer after save model

* fix torch.dequantize_weight

* fix bias

* fix test_export

* add dynamic check

* add convert_gptq_to_mlx_weights

* add backend.mlx

* load backend.mlx

* fix load

* fix group size

* add mlx_generate

* fix generate

* fix load mlx model

* Update loader.py

* Rename test_export.py to test_mlx.py

* Update backend.py

* Revert "Update loader.py"

This reverts commit 1366d35.

* Update setup.py

* Update loader.py

* add mlx check

---------

Co-authored-by: LRL-ModelCloud <[email protected]>
Co-authored-by: CL-ModelCloud <[email protected]>
Co-authored-by: Qubitium-ModelCloud <[email protected]>

* mlx test

* add generate test

* [CI] upload source & use same codes to test

* [CI] disable show-statistics

* [CI] fix dir exists

* [CI] fix dir exists

* [CI] fix hash

* [CI] fix file name

* [CI] print tags

* [CI] print enve

* [CI] fix compress

* [CI] print files name

* [CI] print files name

* [CI] always run build, but skip compile

* [CI] rename step

* [CI] update uploading source

* [CI] install mlx

* [CI] remove mlx

* quantize lm_head

* update

* Fix incorrect call to layer.forward()

* lm_head uses a special quantize config

* remove store_lm_head_input_hook()

* added code of save/load lm_head_layer_inputs.pt

* fix pack_module()

* remove pack_module()

* Check if quant lm_head supports

* cleanup

* add only_quant_lm_head

* fix only_quant_lm_head

* add store_lm_head_input_hook()

* fix lm_head layer forward error with marlin

* Revert "add store_lm_head_input_hook()"

This reverts commit 10c97a8.

* cleanup

* QuantizeConfig add "lm_head_low_gpu_mem_usage" field

* add TestLmHeadQuant

* fix merge error

* revert marlin dequanitze code

* move dequantize_weight -> qlinear.utils

* [CI] move mlx test to m4

* [CI] fix syntax

* [CI] update build if

* [CI] fix mlx-files

* [CI] check not ''

* [CI] update build if

* [CI]  if

* [CI]  if

* [CI] print if env

* [CI] remove alywas

* [CI] remove cancel

* [CI] Print conditions and parameters

* [CI] update outputs

* [CI] use _

* [CI] add needs

* [CI] add needs

* [CI] rename test sh

* [CI] fix parameter not received

* [CI] rename

* [CI] update

* [CI] update if

* [CI] remove ignore

* [CI] clean local

* [CI] clean local

* [CI] update

* darwin BUILD_CUDA_EXT false

* [CI] add test var

* [CI] append .py

* [CI] fix regex

* update prompt

* no need redo quant model

* fix import

* [CI] replace model path

* [CI] replace model path

* [CI] fix path

* [CI] update path

* [CI] update path

* [CI] fix replace

* fix not export

* remove     deprecated repetition_penalty

* check none

* [CI] remove all in clean cache step

* print repo & ref

* use zen3 public ip

* [CI] install with index

* [CI] print dir at top

* update prompt

* import at top

* fix python not activated

* force update

* [CI] deelte hidden files

* [CI] update rm

* [CI] fix clean error

* [CI] fix clean error

* [CI] check monster dir is mounted

* [CI] check monster dir is mounted

* add ModuleNotFoundError

* sys.exit with error msg

* Update setup.py

---------

Co-authored-by: Qubitium-ModelCloud <[email protected]>

* prepare for v1.7.0 release

* Update version.py

* Update README.md