Exception during processing !!! Activation scale shape mismatch: scale.shape=torch.Size([96, 24]), expected (12288, 24)

[FurkanGozukara]

Testing qwen image qwen_image_edit_2511_int8mixed.safetensors

2025-12-23 23:34:32.610 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\custom_nodes\ComfyUI-QuantOps\quant_layouts\int8_layout.py", line 277, in dequantize
2025-12-23 23:34:32.611 [Warning] [ComfyUI-0/STDERR]     raise RuntimeError(
2025-12-23 23:34:32.611 [Warning] [ComfyUI-0/STDERR] RuntimeError: Activation scale shape mismatch: scale.shape=torch.Size([96, 24]), expected (12288, 24)

2025-12-23 23:34:32.587 [Debug] [ComfyUI-0/STDERR] !!! Exception during processing !!! Activation scale shape mismatch: scale.shape=torch.Size([96, 24]), expected (12288, 24)
2025-12-23 23:34:32.593 [Warning] [ComfyUI-0/STDERR] Traceback (most recent call last):
2025-12-23 23:34:32.594 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\execution.py", line 516, in execute
2025-12-23 23:34:32.594 [Warning] [ComfyUI-0/STDERR]     output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
2025-12-23 23:34:32.594 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\execution.py", line 330, in get_output_data
2025-12-23 23:34:32.595 [Warning] [ComfyUI-0/STDERR]     return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
2025-12-23 23:34:32.595 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\execution.py", line 304, in _async_map_node_over_list
2025-12-23 23:34:32.595 [Warning] [ComfyUI-0/STDERR]     await process_inputs(input_dict, i)
2025-12-23 23:34:32.596 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\execution.py", line 292, in process_inputs
2025-12-23 23:34:32.596 [Warning] [ComfyUI-0/STDERR]     result = f(**inputs)
2025-12-23 23:34:32.596 [Warning] [ComfyUI-0/STDERR]   File "F:\SwarmUI_Model_Downloader_v81\SwarmUI\src\BuiltinExtensions\ComfyUIBackend\ExtraNodes\SwarmComfyCommon\SwarmKSampler.py", line 357, in run_sampling
2025-12-23 23:34:32.596 [Warning] [ComfyUI-0/STDERR]     return self.sample(model, noise_seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, start_at_step, end_at_step, var_seed, var_seed_strength, sigma_max, sigma_min, rho, add_noise, return_with_leftover_noise, previews)
2025-12-23 23:34:32.597 [Warning] [ComfyUI-0/STDERR]   File "F:\SwarmUI_Model_Downloader_v81\SwarmUI\src\BuiltinExtensions\ComfyUIBackend\ExtraNodes\SwarmComfyCommon\SwarmKSampler.py", line 331, in sample
2025-12-23 23:34:32.597 [Warning] [ComfyUI-0/STDERR]     samples = comfy.sample.sample(model, noise, steps, cfg, sampler_name, scheduler, positive, negative, latent_samples,
2025-12-23 23:34:32.597 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\sample.py", line 60, in sample
2025-12-23 23:34:32.598 [Warning] [ComfyUI-0/STDERR]     samples = sampler.sample(noise, positive, negative, cfg=cfg, latent_image=latent_image, start_step=start_step, last_step=last_step, force_full_denoise=force_full_denoise, denoise_mask=noise_mask, sigmas=sigmas, callback=callback, disable_pbar=disable_pbar, seed=seed)
2025-12-23 23:34:32.598 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\samplers.py", line 1178, in sample
2025-12-23 23:34:32.598 [Warning] [ComfyUI-0/STDERR]     return sample(self.model, noise, positive, negative, cfg, self.device, sampler, sigmas, self.model_options, latent_image=latent_image, denoise_mask=denoise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed)
2025-12-23 23:34:32.599 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\samplers.py", line 1068, in sample
2025-12-23 23:34:32.599 [Warning] [ComfyUI-0/STDERR]     return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)
2025-12-23 23:34:32.599 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\samplers.py", line 1050, in sample
2025-12-23 23:34:32.600 [Warning] [ComfyUI-0/STDERR]     output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
2025-12-23 23:34:32.600 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\patcher_extension.py", line 112, in execute
2025-12-23 23:34:32.600 [Warning] [ComfyUI-0/STDERR]     return self.original(*args, **kwargs)
2025-12-23 23:34:32.601 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\samplers.py", line 984, in outer_sample
2025-12-23 23:34:32.601 [Warning] [ComfyUI-0/STDERR]     self.inner_model, self.conds, self.loaded_models = comfy.sampler_helpers.prepare_sampling(self.model_patcher, noise.shape, self.conds, self.model_options)
2025-12-23 23:34:32.602 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\sampler_helpers.py", line 130, in prepare_sampling
2025-12-23 23:34:32.602 [Warning] [ComfyUI-0/STDERR]     return executor.execute(model, noise_shape, conds, model_options=model_options, force_full_load=force_full_load)
2025-12-23 23:34:32.602 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\patcher_extension.py", line 112, in execute
2025-12-23 23:34:32.603 [Warning] [ComfyUI-0/STDERR]     return self.original(*args, **kwargs)
2025-12-23 23:34:32.603 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\sampler_helpers.py", line 138, in _prepare_sampling
2025-12-23 23:34:32.603 [Warning] [ComfyUI-0/STDERR]     comfy.model_management.load_models_gpu([model] + models, memory_required=memory_required + inference_memory, minimum_memory_required=minimum_memory_required + inference_memory, force_full_load=force_full_load)
2025-12-23 23:34:32.604 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\model_management.py", line 704, in load_models_gpu
2025-12-23 23:34:32.604 [Warning] [ComfyUI-0/STDERR]     loaded_model.model_load(lowvram_model_memory, force_patch_weights=force_patch_weights)
2025-12-23 23:34:32.604 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\model_management.py", line 509, in model_load
2025-12-23 23:34:32.604 [Warning] [ComfyUI-0/STDERR]     self.model_use_more_vram(use_more_vram, force_patch_weights=force_patch_weights)
2025-12-23 23:34:32.605 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\model_management.py", line 539, in model_use_more_vram
2025-12-23 23:34:32.605 [Warning] [ComfyUI-0/STDERR]     return self.model.partially_load(self.device, extra_memory, force_patch_weights=force_patch_weights)
2025-12-23 23:34:32.605 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\model_patcher.py", line 981, in partially_load
2025-12-23 23:34:32.606 [Warning] [ComfyUI-0/STDERR]     raise e
2025-12-23 23:34:32.606 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\model_patcher.py", line 978, in partially_load
2025-12-23 23:34:32.606 [Warning] [ComfyUI-0/STDERR]     self.load(device_to, lowvram_model_memory=current_used + extra_memory, force_patch_weights=force_patch_weights, full_load=full_load)
2025-12-23 23:34:32.607 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\model_patcher.py", line 777, in load
2025-12-23 23:34:32.608 [Warning] [ComfyUI-0/STDERR]     self.patch_weight_to_device(key, device_to=device_to)
2025-12-23 23:34:32.608 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\model_patcher.py", line 630, in patch_weight_to_device
2025-12-23 23:34:32.608 [Warning] [ComfyUI-0/STDERR]     temp_weight = convert_func(temp_weight, inplace=True)
2025-12-23 23:34:32.609 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\ops.py", line 619, in convert_weight
2025-12-23 23:34:32.609 [Warning] [ComfyUI-0/STDERR]     return weight.dequantize()
2025-12-23 23:34:32.610 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\comfy\quant_ops.py", line 197, in dequantize
2025-12-23 23:34:32.610 [Warning] [ComfyUI-0/STDERR]     return LAYOUTS[self._layout_type].dequantize(self._qdata, **self._layout_params)
2025-12-23 23:34:32.610 [Warning] [ComfyUI-0/STDERR]   File "F:\Comfy_UI_V52\ComfyUI\custom_nodes\ComfyUI-QuantOps\quant_layouts\int8_layout.py", line 277, in dequantize
2025-12-23 23:34:32.611 [Warning] [ComfyUI-0/STDERR]     raise RuntimeError(
2025-12-23 23:34:32.611 [Warning] [ComfyUI-0/STDERR] RuntimeError: Activation scale shape mismatch: scale.shape=torch.Size([96, 24]), expected (12288, 24)
2025-12-23 23:34:32.611 [Warning] [ComfyUI-0/STDERR] 
2025-12-23 23:34:32.612 [Debug] [ComfyUI-0/STDERR] Prompt executed in 3.20 seconds
2025-12-23 23:34:32.693 [Debug] Failed to process comfy workflow for inputs T2IParamInput(prompt: turn into pixel art, model: qwen_image_edit_2511_int8mixed, seed: 1652804466, steps: 12, cfgscale: 1, aspectratio: 16:9, width: 1744, height: 992, sampler: euler_ancestral, scheduler: simple, automaticvae: True, loras: Qwen-Image-Edit-2511-Lightning-4steps-V1.0-fp32, loraweights: 1, preferreddtype: default, promptimages: iVBORw0KGgoAAAANSUhEUgAADSAAAAeACAIAAADrN4/EAAAQAElEQVR4AVT9BcAexbU/AJ8zsruPvBLH3d3dXYtbaaG99fbeym2plwLFSSCGExIS3J0gxVqsuAWS4Bp79bHdHft+84T//b5v3nlmZ2fOHD9nZndTKh770D+4MDy0MDz8fnhkYazoPLww3j74XnhwQXjovfDQ++GhxeHhReHhhR710Q/Dyjr/g/DoB2H+h+GRDwIw3Puuu+tNe+ur..., negativeprompt: ) with raw workflow { "4": { "class_type": "UNETLoader", "inputs": { "unet_name": "qwen_image_edit_2511_int8mixed.safetensors", "weight_dtype": "default" } }, "100": { "class_type": "CLIPLoader", "inputs": { "clip_name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", "type": "qwen_image", "device": "default" } }, "101": { "class_type": "VAELoader", "inputs": { "vae_name": "QwenImage\\qwen_image_vae.safetensors" } }, "102": { "class_type": "ModelSamplingAuraFlow", "inputs": { "model": [ "4", 0 ], "shift": 3 } }, "3000": { "class_type": "LoraLoaderModelOnly", "inputs": { "model": [ "102", 0 ], "lora_name": "Qwen-Image-Edit-2511-Lightning-4steps-V1.0-fp32.safetensors", "strength_model": 1 } }, "5": { "class_type": "EmptySD3LatentImage", "inputs": { "batch_size": 1, "height": 992, "width": 1744 } }, "103": { "class_type": "SwarmLoadImageB64", "inputs": { "image_base64": "iVBORw0KGgoAAAANSUhEUgAADSAAAAeACAIAAADrN4/EAAAQAElEQVR4AVT9BcAexbU/AJ8zsruPvBLH3d3dXYtbaaG99fbeym2plwLFSSCGExIS3J0gxVqsuAWS4Bp79bHdHft+84T//b5v3nlmZ2fOHD9nZndTKh770D+4MDy0MDz8fnhkYazoPLww3j74XnhwQXjovfDQ++GhxeHhReHhhR710Q/Dyjr/g/DoB2H+h+GRDwIw3Puuu+tNe..." } }, "104": { "class_type": "ImageScale", "inputs": { "image": [ "103", 0 ], "width": 512, "height": 320, "crop": "disabled", "upscale_method": "lanczos" } }, "6": { "class_type": "TextEncodeQwenImageEditPlus", "inputs": { "clip": [ "100", 0 ], "prompt": "turn into pixel art", "vae": null, "image1": [ "104", 0 ], "image2": null, "image3": null } }, "7": { "class_type": "TextEncodeQwenImageEditPlus", "inputs": { "clip": [ "100", 0 ], "prompt": "", "vae": null, "image1": [ "104", 0 ], "image2": null, "image3": null } }, "105": { "class_type": "ImageScale", "inputs": { "image": [ "103", 0 ], "width": 1344, "height": 768, "crop": "disabled", "upscale_method": "lanczos" } }, "106": { "class_type": "VAEEncode", "inputs": { "vae": [ "101", 0 ], "pixels": [ "105", 0 ] } }, "107": { "class_type": "ReferenceLatent", "inputs": { "conditioning": [ "6", 0 ], "latent": [ "106", 0 ] } }, "108": { "class_type": "ReferenceLatent", "inputs": { "conditioning": [ "7", 0 ], "latent": [ "106", 0 ] } }, "10": { "class_type": "SwarmKSampler", "inputs": { "model": [ "3000", 0 ], "noise_seed": 1652804466, "steps": 12, "cfg": 1, "sampler_name": "euler_ancestral", "scheduler": "simple", "positive": [ "107", 0 ], "negative": [ "108", 0 ], "latent_image": [ "5", 0 ], "start_at_step": 0, "end_at_step": 10000, "return_with_leftover_noise": "disable", "add_noise": "enable", "control_after_generate": "fixed", "var_seed": 0, "var_seed_strength": 0, "sigma_min": -1, "sigma_max": -1, "rho": 7, "previews": "default", "tile_sample": False, "tile_size": 1328 } }, "8": { "class_type": "VAEDecode", "inputs": { "vae": [ "101", 0 ], "samples": [ "10", 0 ] } }, "9": { "class_type": "SwarmSaveImageWS", "inputs": { "images": [ "8", 0 ], "bit_depth": "8bit" } } }
2025-12-23 23:34:32.694 [Debug] Refused to generate image for local: ComfyUI execution error: Activation scale shape mismatch: scale.shape=torch.Size([96, 24]), expected (12288, 24)
2025-12-23 23:34:37.012 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 20/114
2025-12-23 23:34:41.941 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 25/114
2025-12-23 23:34:46.049 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 30/114
2025-12-23 23:34:50.326 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 35/114
2025-12-23 23:34:54.504 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 40/114
2025-12-23 23:34:58.666 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 45/114
2025-12-23 23:35:02.845 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 50/114
2025-12-23 23:35:06.866 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 55/114
2025-12-23 23:35:11.105 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 60/114
2025-12-23 23:35:15.797 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 65/114
2025-12-23 23:35:19.901 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 70/114
2025-12-23 23:35:24.221 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 75/114
2025-12-23 23:35:28.910 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 80/114
2025-12-23 23:35:33.407 [Debug] [ComfyUI-0/STDOUT] FETCH ComfyRegistry Data: 85/114

[silveroxides]

@FurkanGozukara Was this resolved yet? I would recommend sticking to for models at the moment though. int8 part still need some work done.

[FurkanGozukara]

@silveroxides nope not fixed

i gave up on int8 all my attempts always fails

also yesterday whole day spent to make a working qwen image 2512 model

the commands i tested on this repo like these generated noise

convert_to_quant -i qwen_image_2512_bf16.safetensors -o outputs\qwen_image_2512_fp8_scaled_tensor.safetensors --comfy_quant --qwen --num_iter 3000 --full_matrix --lr_schedule plateau

convert_to_quant -i qwen_image_2512_bf16.safetensors -o outputs\qwen_image_2512_fp8_scaled_simple_input_scale_no_qwen.safetensors --comfy_quant --simple --input_scale

the below one not even supported

convert_to_quant -i qwen_image_2512_bf16.safetensors -o outputs\qwen_image_2512_fp8_scaled.safetensors --comfy_quant --qwen --num_iter 2000 --full_matrix --lr_schedule plateau --scaling_mode block --block_size 64

convert_to_quant -i qwen_image_2512_bf16.safetensors -o outputs\qwen_image_2512_fp8_scaled_simple_input_scale_no_qwen.safetensors

after that i added code to my app and with below settings finally i got a good model

[Ph0rk0z]

I am able to inference chroma int8 on turning/ampere. The weights simply had to be made from BF16/FP16. Also got hit by relative path bug.

Problem is that I can't compile int8 with triton on this implementation. The quality is better than FP8 but the speed is a whole 3 seconds more. Patched triton on non-mixed FP8 compiles and goes brrr.

For all of us pre-ada people it would neat to have proper quantizer compatible with standard comfy nodes. And yes, it's faster than GGUF, especially if you factor in lora.