Hello, author.
I would like to inquire about how the act_stale file of qwen3 was generated when using the smoothquant method to quantify qwen3?
Is it the generate_act_stcales. py file that is being executed? So may I ask:
--Dataset path: Calibrate the dataset path (default "dataset/val. json. zst")
--Num samples: number of samples (default 512)
--Seq len: Sequence length (default 512)
How were these three command-line parameters set at the time?
The calibration dataset used is eval_my/ppl_datasets/alenai/c4/rain
Is this' eval_my/ppl_datasets/alenai/c4/validation '? I see that there are many datasets (en.noblocklist, en.noclean, en, multilingual,...) in/alenai -- c4. Which one are they located in? Taking en as an example, there are many datasets (c4-train. 00000-of-01024. json. gz, c4-validation. 00000-of-000008. json. gz) inside. Which dataset is used specifically?
Hello, author.
I would like to inquire about how the act_stale file of qwen3 was generated when using the smoothquant method to quantify qwen3?
Is it the generate_act_stcales. py file that is being executed? So may I ask:
--Dataset path: Calibrate the dataset path (default "dataset/val. json. zst")
--Num samples: number of samples (default 512)
--Seq len: Sequence length (default 512)
How were these three command-line parameters set at the time?
The calibration dataset used is eval_my/ppl_datasets/alenai/c4/rain
Is this' eval_my/ppl_datasets/alenai/c4/validation '? I see that there are many datasets (en.noblocklist, en.noclean, en, multilingual,...) in/alenai -- c4. Which one are they located in? Taking en as an example, there are many datasets (c4-train. 00000-of-01024. json. gz, c4-validation. 00000-of-000008. json. gz) inside. Which dataset is used specifically?