Create README.md

README.md

@@ -0,0 +1,509 @@
+---
+license: apache-2.0
+pipeline_tag: text-generation
+tags:
+- fp8
+- quantized
+- llm-compressor
+- compressed-tensors
+- red hat
+base_model:
+- ibm-granite/granite-4.0-h-small
+---
+
+
+# Granite-4.0-h-small
+
+## Model Overview
+- **Model Architecture:** GraniteMoeHybridForCausalLM
+  - **Input:** Text
+  - **Output:** Text
+- **Model Optimizations:**
+  - **Weight quantization:** FP8
+  - **Activation quantization:** FP8
+- **Release Date:** 
+- **Version:** 1.0
+- **Model Developers:**: Red Hat
+
+Quantized version of [ibm-granite/granite-4.0-h-small](https://huggingface.co/ibm-granite/granite-4.0-h-small).
+
+### Model Optimizations
+
+This model was obtained by quantizing the weights and activations of [ibm-granite/granite-4.0-h-small](https://huggingface.co/ibm-granite/granite-4.0-h-small) to FP8 data type.
+This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%.
+Only the weights and activations of the linear operators within transformers blocks of the language model are quantized. 
+
+## Deployment
+
+### Use with vLLM
+
+1. Initialize vLLM server:
+```
+vllm serve RedHatAI/granite-4.0-h-small-FP8-block --tensor_parallel_size 4
+```
+
+2. Send requests to the server:
+
+```python
+from openai import OpenAI
+
+# Modify OpenAI's API key and API base to use vLLM's API server.
+openai_api_key = "EMPTY"
+openai_api_base = "http://<your-server-host>:8000/v1"
+
+client = OpenAI(
+    api_key=openai_api_key,
+    base_url=openai_api_base,
+)
+
+model = "RedHatAI/granite-4.0-h-small-FP8-block"
+
+messages = [
+    {"role": "user", "content": "Explain quantum mechanics clearly and concisely."},
+]
+
+
+outputs = client.chat.completions.create(
+    model=model,
+    messages=messages,
+)
+
+generated_text = outputs.choices[0].message.content
+print(generated_text)
+```
+
+<!-- ## Creation
+
+This model was quantized using the [llm-compressor](https://github.com/vllm-project/llm-compressor) library as shown below.
+
+<details>
+  <summary>Creation details</summary>
+
+```python
+from transformers import AutoProcessor, Qwen3ForCausalLM
+
+from llmcompressor import oneshot
+from llmcompressor.modeling import replace_modules_for_calibration
+from llmcompressor.modifiers.quantization import QuantizationModifier
+
+MODEL_ID = "Qwen/Qwen3-8B"
+
+# Load model.
+model = Qwen3ForCausalLM.from_pretrained(MODEL_ID, dtype="auto")
+processor = AutoProcessor.from_pretrained(MODEL_ID)
+model = replace_modules_for_calibration(model)
+
+# Configure the quantization algorithm and scheme.
+# In this case, we:
+#   * quantize the weights to fp8 with per-block quantization
+#   * quantize the activations to fp8 with dynamic token activations
+recipe = QuantizationModifier(
+    targets="Linear",
+    scheme="FP8_BLOCK",
+    ignore=["lm_head"],
+)
+
+# Apply quantization.
+oneshot(model=model, recipe=recipe)
+
+# Save to disk in compressed-tensors format.
+SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-FP8-block"
+model.save_pretrained(SAVE_DIR)
+processor.save_pretrained(SAVE_DIR)
+```
+</details> -->
+
+
+## Evaluation
+
+
+The model was evaluated on the OpenLLM leaderboard task, using [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness).
+[vLLM](https://docs.vllm.ai/en/stable/) was used for all evaluations.
+
+<details>
+  <summary>Evaluation details</summary>
+  
+  **Openllm V1**
+  ```
+  lm_eval \
+    --model vllm \
+    --model_args pretrained="RedHatAI/granite-4.0-h-small-FP8-block",dtype=auto,add_bos_token=True,max_model_len=16384,tensor_parallel_size=4,gpu_memory_utilization=0.9,enable_chunked_prefill=True,trust_remote_code=True \
+    --tasks openllm \
+    --write_out \
+    --batch_size auto \
+    --show_config
+  ```
+
+
+  **Openllm V2**  
+  ```
+  lm_eval \
+    --model vllm \
+    --model_args pretrained="RedHatAI/granite-4.0-h-small-FP8-block",dtype=auto,add_bos_token=False,max_model_len=16384,tensor_parallel_size=4,gpu_memory_utilization=0.7,disable_log_stats=True,enable_chunked_prefill=True,trust_remote_code=True \
+    --tasks leaderboard \
+    --apply_chat_template \
+    --fewshot_as_multiturn \
+    --write_out \
+    --batch_size auto \
+    --show_config
+  ```
+
+
+  **Coding Benchmarks**
+
+  ```
+  evalplus.evaluate --model "RedHatAI/granite-4.0-h-small-FP8-block" \
+                    --dataset "humaneval" \
+                    --backend vllm \
+                    --tp 4 \
+                    --greedy
+
+  evalplus.evaluate --model "RedHatAI/granite-4.0-h-small-FP8-block" \
+                  --dataset "mbpp" \
+                  --backend vllm \
+                  --tp 4 \
+                  --greedy
+
+  ```
+
+</details>
+
+
+<b>*</b> I/p Length = 2048, O/p Length = 2048, #Requests = 1024
+
+### Accuracy
+<table>
+  
+  <thead>
+    <tr>
+      <th>Category</th>
+      <th>Metric</th>
+      <th>ibm-granite/granite-4.0-h-small</th>
+      <th>ibm-granite/granite-4.0-h-small-FP8</th>
+      <th>RedHatAI/granite-4.0-h-small-FP8-block</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td rowspan="1"><b>Model Size (GB)</b></td>
+      <td></td>
+      <td>64.41</td>
+      <td>33.48</td>
+      <td>36.43</td>
+    </tr>
+    <tr>
+      <td rowspan="1"><b>Throughput (Requests/sec)*</b></td>
+      <td></td>
+      <td>2.031</td>
+      <td>2.144</td>
+      <td>2.066</td>
+    </tr>
+    <tr>
+      <td rowspan="7"><b>OpenLLM V1</b></td>
+      <td>ARC-Challenge (Acc-Norm, 25-shot)</td>
+      <td>72.27</td>
+      <td>71.67 (99.17)</td>
+      <td>72.27 (100.00)</td>
+    </tr>
+    <tr>
+      <td>GSM8K (Strict-Match, 5-shot)</td>
+      <td>85.06</td>
+      <td>85.60 (100.62)</td>
+      <td>85.60 (100.62)</td>
+    </tr>
+    <tr>
+      <td>HellaSwag (Acc-Norm, 10-shot)</td>
+      <td>86.07</td>
+      <td>86.02 (99.94)</td>
+      <td>85.96 (99.87)</td>
+    </tr>
+    <tr>
+      <td>MMLU (Acc, 5-shot)</td>
+      <td>77.15</td>
+      <td>76.94 (99.73)</td>
+      <td>77.23 (100.10)</td>
+    </tr>
+    <tr>
+      <td>TruthfulQA (MC2, 0-shot)</td>
+      <td>57.97</td>
+      <td>57.62 (99.40)</td>
+      <td>57.85 (99.80)</td>
+    </tr>
+    <tr>
+      <td>Winogrande (Acc, 5-shot)</td>
+      <td>81.45</td>
+      <td>81.14 (99.61)</td>
+      <td>80.82 (99.22)</td>
+    </tr>
+    <tr>
+      <td><b>Average Score</b></td>
+      <td><b>76.66</b></td>
+      <td><b>76.50 (99.79)</b></td>
+      <td><b>76.62 (99.95)</b></td>
+    </tr>
+    <!-- OpenLLM Leaderboard V2 -->
+    <tr>
+      <td rowspan="7"><b>OpenLLM V2</b></td>
+      <td>IFEval (Inst Level Strict Acc, 0-shot)</td>
+      <td>87.41</td>
+      <td>87.65 (100.27)</td>
+      <td>87.89 (100.55)</td>
+    </tr>
+    <tr>
+      <td>BBH (Acc-Norm, 3-shot)</td>
+      <td>61.52</td>
+      <td>61.31 (99.66)</td>
+      <td>61.40 (99.80)</td>
+    </tr>
+    <tr>
+      <td>Math-Hard (Exact-Match, 4-shot)</td>
+      <td>46.60</td>
+      <td>44.34 (95.14)</td>
+      <td>44.94 (96.43)</td>
+    </tr>
+    <tr>
+      <td>GPQA (Acc-Norm, 0-shot)</td>
+      <td>32.55</td>
+      <td>32.05 (98.45)</td>
+      <td>34.23 (105.15)</td>
+    </tr>
+    <tr>
+      <td>MUSR (Acc-Norm, 0-shot)</td>
+      <td>46.43</td>
+      <td>46.30 (99.72)</td>
+      <td>45.77 (98.58)</td>
+    </tr>
+    <tr>
+      <td>MMLU-Pro (Acc, 5-shot)</td>
+      <td>47.96</td>
+      <td>47.91 (99.88)</td>
+      <td>47.93 (99.93)</td>
+    </tr>
+    <tr>
+      <td><b>Average Score</b></td>
+      <td><b>53.75</b></td>
+      <td><b>53.26 (99.09)</b></td>
+      <td><b>53.69 (99.89)</b></td>
+    </tr>
+  </tbody>
+</table>
+
+
+
+
+
+
+<!-- 
+### Accuracy
+<table>
+  <thead>
+    <tr>
+      <th>Category</th>
+      <th>Metric</th>
+      <th>ibm-granite/granite-4.0-h-small</th>
+      <th>ibm-granite/granite-4.0-h-small-FP8</th>
+      <th>Recovery (%)</th>
+    </tr>
+  </thead>
+  <tbody>
+    
+    <tr>
+      <td rowspan="7"><b>OpenLLM V1</b></td>
+      <td>ARC-Challenge (Acc-Norm, 25-shot)</td>
+      <td>72.27</td>
+      <td>71.67</td>
+      <td>99.17</td>
+    </tr>
+    <tr>
+      <td>GSM8K (Strict-Match, 5-shot)</td>
+      <td>85.06</td>
+      <td>85.60</td>
+      <td>100.62</td>
+    </tr>
+    <tr>
+      <td>HellaSwag (Acc-Norm, 10-shot)</td>
+      <td>86.07</td>
+      <td>86.02</td>
+      <td>99.94</td>
+    </tr>
+    <tr>
+      <td>MMLU (Acc, 5-shot)</td>
+      <td>77.15</td>
+      <td>76.94</td>
+      <td>99.73</td>
+    </tr>
+    <tr>
+      <td>TruthfulQA (MC2, 0-shot)</td>
+      <td>57.97</td>
+      <td>57.62</td>
+      <td>99.40</td>
+    </tr>
+    <tr>
+      <td>Winogrande (Acc, 5-shot)</td>
+      <td>81.45</td>
+      <td>81.14</td>
+      <td>99.61</td>
+    </tr>
+    <tr>
+      <td><b>Average Score</b></td>
+      <td><b>76.66</b></td>
+      <td><b>76.50</b></td>
+      <td><b>99.79</b></td>
+    </tr>
+   
+    <tr>
+      <td rowspan="7"><b>OpenLLM V2</b></td>
+      <td>IFEval (Inst Level Strict Acc, 0-shot)</td>
+      <td>87.41</td>
+      <td>87.65</td>
+      <td>100.27</td>
+    </tr>
+    <tr>
+      <td>BBH (Acc-Norm, 3-shot)</td>
+      <td>61.52</td>
+      <td>61.31</td>
+      <td>99.66</td>
+    </tr>
+    <tr>
+      <td>Math-Hard (Exact-Match, 4-shot)</td>
+      <td>46.60</td>
+      <td>44.34</td>
+      <td>95.14</td>
+    </tr>
+    <tr>
+      <td>GPQA (Acc-Norm, 0-shot)</td>
+      <td>32.55</td>
+      <td>32.05</td>
+      <td>98.45</td>
+    </tr>
+    <tr>
+      <td>MUSR (Acc-Norm, 0-shot)</td>
+      <td>46.43</td>
+      <td>46.30</td>
+      <td>99.72</td>
+    </tr>
+    <tr>
+      <td>MMLU-Pro (Acc, 5-shot)</td>
+      <td>47.96</td>
+      <td>47.91</td>
+      <td>99.88</td>
+    </tr>
+    <tr>
+      <td><b>Average Score</b></td>
+      <td><b>53.75</b></td>
+      <td><b>53.26</b></td>
+      <td><b>99.09</b></td>
+    </tr>
+  </tbody>
+</table>
+
+
+
+
+
+
+
+
+
+### Accuracy
+<table>
+  <thead>
+    <tr>
+      <th>Category</th>
+      <th>Metric</th>
+      <th>ibm-granite/granite-4.0-h-small</th>
+      <th>RedHatAI/granite-4.0-h-small-FP8-block</th>
+      <th>Recovery (%)</th>
+    </tr>
+  </thead>
+  <tbody>
+    
+    <tr>
+      <td rowspan="7"><b>OpenLLM V1</b></td>
+      <td>ARC-Challenge (Acc-Norm, 25-shot)</td>
+      <td>72.27</td>
+      <td>72.27</td>
+      <td>100.00</td>
+    </tr>
+    <tr>
+      <td>GSM8K (Strict-Match, 5-shot)</td>
+      <td>85.06</td>
+      <td>85.60</td>
+      <td>100.62</td>
+    </tr>
+    <tr>
+      <td>HellaSwag (Acc-Norm, 10-shot)</td>
+      <td>86.07</td>
+      <td>85.96</td>
+      <td>99.87</td>
+    </tr>
+    <tr>
+      <td>MMLU (Acc, 5-shot)</td>
+      <td>77.15</td>
+      <td>77.23</td>
+      <td>100.10</td>
+    </tr>
+    <tr>
+      <td>TruthfulQA (MC2, 0-shot)</td>
+      <td>57.97</td>
+      <td>57.85</td>
+      <td>99.80</td>
+    </tr>
+    <tr>
+      <td>Winogrande (Acc, 5-shot)</td>
+      <td>81.45</td>
+      <td>80.82</td>
+      <td>99.22</td>
+    </tr>
+    <tr>
+      <td><b>Average Score</b></td>
+      <td><b>76.66</b></td>
+      <td><b>76.62</b></td>
+      <td><b>99.95</b></td>
+    </tr>
+    
+    <tr>
+      <td rowspan="7"><b>OpenLLM V2</b></td>
+      <td>IFEval (Inst Level Strict Acc, 0-shot)</td>
+      <td>87.41</td>
+      <td>87.89</td>
+      <td>100.55</td>
+    </tr>
+    <tr>
+      <td>BBH (Acc-Norm, 3-shot)</td>
+      <td>61.52</td>
+      <td>61.40</td>
+      <td>99.80</td>
+    </tr>
+    <tr>
+      <td>Math-Hard (Exact-Match, 4-shot)</td>
+      <td>46.60</td>
+      <td>44.94</td>
+      <td>96.43</td>
+    </tr>
+    <tr>
+      <td>GPQA (Acc-Norm, 0-shot)</td>
+      <td>32.55</td>
+      <td>34.23</td>
+      <td>105.15</td>
+    </tr>
+    <tr>
+      <td>MUSR (Acc-Norm, 0-shot)</td>
+      <td>46.43</td>
+      <td>45.77</td>
+      <td>98.58</td>
+    </tr>
+    <tr>
+      <td>MMLU-Pro (Acc, 5-shot)</td>
+      <td>47.96</td>
+      <td>47.93</td>
+      <td>99.93</td>
+    </tr>
+    <tr>
+      <td><b>Average Score</b></td>
+      <td><b>53.75</b></td>
+      <td><b>53.69</b></td>
+      <td><b>99.89</b></td>
+    </tr>
+  </tbody>
+</table> -->