granite-4.0-h-small-FP8-block

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.

Model Optimizations

This model was obtained by quantizing the weights and activations of 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. Install specific version:

uv pip install -U git+https://github.com/vllm-project/vllm.git \
  --extra-index-url https://wheels.vllm.ai/nightly \
  --no-deps \
  --no-cache

uv pip install compressed-tensors==0.12.3a20251114 --no-cache
uv pip install --upgrade torchvision --break-system-packages --no-cache
uv pip install cloudpickle msgspec zmq blake3 cachetools prometheus_client fastapi openai openai_harmony pybase64 llguidance diskcache xgrammar lm-format-enforcer partial-json-parser cbor2 einops gguf numba --no-cache

1. Initialize vLLM server:

vllm serve RedHatAI/granite-4.0-h-small-FP8-block --tensor_parallel_size 1

1. Send requests to the server:

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 library as shown below.

<details> <summary>Creation details</summary>

Install specific llm-compression version:

uv pip install git+https://github.com/vllm-project/llm-compressor.git@refs/pull/2001/head --no-cache
uv pip install --upgrade torchvision --break-system-packages --no-cache

from transformers import AutoModelForCausalLM, AutoTokenizer

from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.utils import dispatch_for_generation
from llmcompressor.modeling import replace_modules_for_calibration
from llmcompressor.modeling.granite4 import pack_3d_experts

MODEL_ID = "ibm-granite/granite-4.0-h-small"

model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

model = replace_modules_for_calibration(model)

ignore_lay = ["lm_head", "re:.*block_sparse_moe.router", "re:.*mamba.in_proj", "re:.*shared_mlp.input_linear"]

recipe = QuantizationModifier(
    targets=["Linear"],
    scheme="FP8_BLOCK",
    ignore=ignore_lay,
)

oneshot(model=model, recipe=recipe)

print("========== SAMPLE GENERATION ==============")
dispatch_for_generation(model)
input_ids = tokenizer(
    "Describe Large Language Model", return_tensors="pt"
).input_ids.to(model.device)
output = model.generate(input_ids, max_new_tokens=35)
print(tokenizer.decode(output[0]))
print("==========================================")

SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-FP8-block"
print(f"Saving to {SAVE_DIR}")

model.save_pretrained(SAVE_DIR)
tokenizer.save_pretrained(SAVE_DIR)
pack_3d_experts(SAVE_DIR)

</details>

Evaluation

The model was evaluated on the OpenLLM leaderboard task, using lm-evaluation-harness. vLLM was used for all evaluations.

<details> <summary>Evaluation details</summary>

Install specific version:

uv pip install -U git+https://github.com/vllm-project/vllm.git \
  --extra-index-url https://wheels.vllm.ai/nightly \
  --no-deps \
  --no-cache


uv pip install compressed-tensors==0.12.3a20251114 --no-cache
uv pip install --upgrade torchvision --break-system-packages --no-cache
uv pip install cloudpickle msgspec zmq blake3 cachetools prometheus_client fastapi openai openai_harmony pybase64 llguidance diskcache xgrammar lm-format-enforcer partial-json-parser cbor2 einops gguf numba --no-cache

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=1,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=1,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 1 \
                  --greedy

evalplus.evaluate --model "RedHatAI/granite-4.0-h-small-FP8-block" \
                --dataset "mbpp" \
                --backend vllm \
                --tp 1 \
                --greedy

</details> <!-- <b>*</b> I/p Length = 2048, O/p Length = 2048, #Requests = 1024 -->

Accuracy Comparison

<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> <th>RedHatAI/granite-4.0-h-small-FP8-dynamic</th> </tr> </thead> <tbody> <!-- OpenLLM Leaderboard V1 --> <tr> <td rowspan="7"><b>OpenLLM V1</b></td> <td>ARC-Challenge (Acc-Norm, 25-shot)</td> <td>72.27</td> <td>72.10 (99.76%)</td> <td>72.27 (100.00%)</td> <td>72.10 (99.76%)</td> </tr> <tr> <td>GSM8K (Strict-Match, 5-shot)</td> <td>85.22</td> <td>85.29 (100.09%)</td> <td>85.52 (100.36%)</td> <td>84.84 (99.56%)</td> </tr> <tr> <td>HellaSwag (Acc-Norm, 10-shot)</td> <td>86.08</td> <td>85.88 (99.77%)</td> <td>85.96 (99.86%)</td> <td>85.88 (99.77%)</td> </tr> <tr> <td>MMLU (Acc, 5-shot)</td> <td>77.15</td> <td>77.18 (100.03%)</td> <td>77.23 (100.09%)</td> <td>77.18 (100.03%)</td> </tr> <tr> <td>TruthfulQA (MC2, 0-shot)</td> <td>57.64</td> <td>57.63 (99.99%)</td> <td>57.94 (100.52%)</td> <td>57.63 (100.00%)</td> </tr> <tr> <td>Winogrande (Acc, 5-shot)</td> <td>81.37</td> <td>81.45 (100.10%)</td> <td>80.82 (99.32%)</td> <td>81.45 (100.10%)</td> </tr> <tr> <td><b>Average Score</b></td> <td><b>76.62</b></td> <td><b>76.59 (99.96%)</b></td> <td><b>76.62 (100.00%)</b></td> <td><b>76.51 (99.86%)</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>87.53</td> <td>87.17 (99.59%)</td> <td>86.69 (99.04%)</td> <td>87.41 (99.86%)</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> <td>61.19 (99.46%)</td> </tr> <tr> <td>Math-Hard (Exact-Match, 4-shot)</td> <td>46.22</td> <td>43.73 (94.61%)</td> <td>43.88 (94.93%)</td> <td>41.77 (90.36%)</td> </tr> <tr> <td>GPQA (Acc-Norm, 0-shot)</td> <td>35.23</td> <td>34.98 (99.29%)</td> <td>34.23 (97.14%)</td> <td>34.23 (97.14%)</td> </tr> <tr> <td>MUSR (Acc-Norm, 0-shot)</td> <td>46.69</td> <td>46.56 (99.72%)</td> <td>45.77 (98.02%)</td> <td>45.77 (98.02%)</td> </tr> <tr> <td>MMLU-Pro (Acc, 5-shot)</td> <td>47.99</td> <td>47.63 (99.26%)</td> <td>47.93 (99.88%)</td> <td>47.58 (99.15%)</td> </tr> <tr> <td><b>Average Score</b></td> <td><b>54.20</b></td> <td><b>53.56 (98.82%)</b></td> <td><b>53.32 (98.38%)</b></td> <td><b>52.99 (97.77%)</b></td> </tr> </tbody> </table>