Upload folder using huggingface_hub

README.md

@@ -34,10 +34,10 @@ The model utilizes **static FP8 quantization** for optimal inference performance
 
 - **Original Model**: [HuggingFaceTB/SmolLM-135M](https://huggingface.co/HuggingFaceTB/SmolLM-135M)
 - **Source Model**: HuggingFaceTB/SmolLM-135M
-- **Quantized Model**: InternVL3-38B-FP8-Dynamic
-- **Quantization Method**: FP8 Dynamic (W8A8) 
-- **Quantization Library**: [LLM Compressor](https://github.com/vllm-project/llm-compressor) v0.6.0
-- **Calibration Dataset**: N/A
+- **Quantized Model**: InternVL3-38B-FP8-Static
+- **Quantization Method**: FP8 Static (W8A8) 
+- **Quantization Library**: [LLM Compressor](https://github.com/vllm-project/llm-compressor) v0.6.1.dev18+g090baff5
+- **Calibration Dataset**: open_platypus (256 samples, seq_len=2048)
 - **Attention Implementation**: Flash Attention 2 (memory efficient, fastest)
 - **Quantized by**: [JustJaro](https://huggingface.co/JustJaro)
 
@@ -50,7 +50,7 @@ from vllm import LLM, SamplingParams
 
 # Load the quantized model
 model = LLM(
-    model="JustJaro/InternVL3-38B-FP8-Dynamic",
+    model="JustJaro/InternVL3-38B-FP8-Static",
     trust_remote_code=True,
     max_model_len=8192,
     tensor_parallel_size=1,  # Adjust based on your GPU setup
@@ -68,7 +68,7 @@ print(response[0].outputs[0].text)
 from transformers import AutoTokenizer, AutoProcessor
 from llmcompressor import LLM
 
-model_id = "JustJaro/InternVL3-38B-FP8-Dynamic"
+model_id = "JustJaro/InternVL3-38B-FP8-Static"
 model = LLM.load(model_id, device="cuda")
 tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
 processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
@@ -93,7 +93,7 @@ print(response)
 - **Weights**: FP8 E4M3 with static per-tensor scales
 - **Activations**: FP8 E4M3 with static per-tensor scales  
 - **Preserved Components**: Vision tower, embeddings, normalization layers
-- **Calibration**: 0 samples from multimodal dataset
+- **Calibration**: 256 samples from multimodal dataset
 
 ## 📈 Performance Benchmarks
 
@@ -109,8 +109,8 @@ Expected performance improvements over FP16 baseline:
 This model was created using:
 
 ```
-llmcompressor==0.6.0
-transformers==4.53.0
+llmcompressor==0.6.1.dev18+g090baff5
+transformers==4.52.4
 torch==2.7.1
 vllm==not installed
 ```
@@ -1031,7 +1031,7 @@ Original model: [HuggingFaceTB/SmolLM-135M](https://huggingface.co/HuggingFaceTB
 ## 📞 Contact
 
 For questions about this quantized model:
-- **Issues**: [Create an issue](https://huggingface.co/JustJaro/InternVL3-38B-FP8-Dynamic/discussions)
+- **Issues**: [Create an issue](https://huggingface.co/JustJaro/InternVL3-38B-FP8-Static/discussions)
 - **Original Model**: Refer to [HuggingFaceTB/SmolLM-135M](https://huggingface.co/HuggingFaceTB/SmolLM-135M)
 
 ---

config.json

@@ -24,12 +24,12 @@
         "input_activations": {
           "actorder": null,
           "block_structure": null,
-          "dynamic": true,
+          "dynamic": false,
           "group_size": null,
           "num_bits": 8,
-          "observer": null,
+          "observer": "minmax",
           "observer_kwargs": {},
-          "strategy": "token",
+          "strategy": "tensor",
           "symmetric": true,
           "type": "float"
         },
@@ -45,7 +45,7 @@
           "num_bits": 8,
           "observer": "minmax",
           "observer_kwargs": {},
-          "strategy": "channel",
+          "strategy": "tensor",
           "symmetric": true,
           "type": "float"
         }
@@ -65,7 +65,7 @@
   "rope_theta": 10000.0,
   "tie_word_embeddings": true,
   "torch_dtype": "bfloat16",
-  "transformers_version": "4.53.0",
+  "transformers_version": "4.52.4",
   "use_cache": true,
   "vocab_size": 49152
 }

generation_config.json

@@ -2,5 +2,5 @@
   "_from_model_config": true,
   "bos_token_id": 0,
   "eos_token_id": 0,
-  "transformers_version": "4.53.0"
+  "transformers_version": "4.52.4"
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b29852221e8b0fb7ce5364816d029fc8cd9fbdc0b790efad61cc32ce4dc2f36
-size 163227736
+oid sha256:071bb4b9ad3b3f7ea5aafeb481e5b5d7f486df3c81006fd643b651328a1367ea
+size 219563872

recipe.yaml

@@ -4,4 +4,4 @@ default_stage:
       targets: [Linear]
       ignore: ['re:.*lm_head', 're:.*vision.*', 're:.*visual.*', 're:.*image.*', 're:.*patch_embed.*',
         're:.*pos_embed.*', 're:.*norm.*', 're:.*layernorm.*']
-      scheme: FP8_DYNAMIC
+      scheme: FP8