Update on "Add FP8-INT4 checkpoint upload code"

Summary:
att, the support is added in #3714
checkpoint: https://huggingface.co/jerryzh168/Qwen3-8B-FP8-INT4

Test Plan:
```
sh release.sh --model_id $MODEL --push_to_hub --populate_model_card_template --quants FP8-INT4
```

produced checkpoint: https://huggingface.co/jerryzh168/Qwen3-8B-FP8-INT4


Benchmark:
```
vllm bench throughput --model jerryzh168/Qwen3-8B-FP8-INT4
```
```
Throughput: 33.03 requests/s, 38055.86 total tokens/s, 4228.43 output tokens/s
Total num prompt tokens:  1024000
Total num output tokens:  128000
```

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]

Author: jerryzh168

.github/scripts/torchao_model_releases/quantize_and_upload.py

@@ -19,7 +19,6 @@
 
 _huggingface_hub_version = str(huggingface_hub.__version__)
 
-from torchao._models._eval import TransformerEvalWrapper
 from torchao.prototype.awq import (
     AWQConfig,
 )
@@ -783,6 +782,8 @@ def filter_fn_skip_lmhead(module, fqn):
         else:
             quantize_(model, awq_config)
 
+        from torchao._models._eval import TransformerEvalWrapper
+
         TransformerEvalWrapper(
             model=model,
             tokenizer=tokenizer,

torchao/quantization/quant_api.py

@@ -877,7 +877,7 @@ def _float8_dynamic_activation_int4_weight_transform(
     )
     weight = module.weight
     group_size = 128
-    block_size = tuple([1 for _ in range(weight.ndim - 1)] + [group_size])
+    block_size = list([1 for _ in range(weight.ndim - 1)] + [group_size])
 
     if int4_packing_format == "preshuffled":
         new_weight = Int4PreshuffledTensor.from_hp(