nm-testing/Llama-3.1-8B-Instruct-KV-Cache-FP8

Llama-3.1-8B-Instruct-KV-Cache-FP8

Model Overview

• —Model Architecture: nm-testing/Llama-3.1-8B-Instruct-KV-Cache-FP8
• —Input: Text
• —Output: Text
• —Release Date:
• —Version: 1.0
• —Model Developers:: Red Hat

FP8 KV Cache Quantization of meta-llama/Llama-3.1-8B-Instruct.

Model Optimizations

This model was obtained by quantizing the KV Cache of weights and activations of meta-llama/Llama-3.1-8B-Instruct to FP8 data type.

Deployment

Use with vLLM

1. 1.Initialize vLLM server:

vllm serve RedHatAI/Llama-3.1-8B-Instruct-KV-Cache-FP8 --tensor_parallel_size 1

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/Llama-3.1-8B-Instruct-KV-Cache-FP8"

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>

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 RULER and long-context benchmarks (LongBench), using lm-evaluation-harness. vLLM was used for all evaluations.

Accuracy

<table> <thead> <tr> <th>Category</th> <th>Metric</th> <th>meta-llama/Llama-3.1-8B-Instruct</th> <th>nm-testing/Llama-3.1-8B-Instruct-KV-Cache-FP8</th> <th>Recovery (%)</th> </tr> </thead> <tbody> <tr> <td rowspan="1"><b>LongBench V1</b></td> <td>Task 1</td> <td>abc</td> <td>ijk</td> <td>xyz</td> </tr> <tr> <td rowspan="6"><b>NIAH</b></td> <td>niahsingle1</td> <td>abc</td> <td>ijk</td> <td>xyz</td> </tr> <tr> <td>niahsingle2</td> <td>abc</td> <td>ijk</td> <td>xyz</td> </tr> <tr> <td>niahsingle3</td> <td>abc</td> <td>ijk</td> <td>xyz</td> </tr> <tr> <td>niahmultikey1</td> <td>abc</td> <td>ijk</td> <td>xyz</td> </tr> <tr> <td>niahmultikey2</td> <td>abc</td> <td>ijk</td> <td>xyz</td> </tr> <tr> <td>niahmultikey3</td> <td>abc</td> <td>ijk</td> <td>xyz</td> </tr> <tr> <td><b>Average Score</b></td> <td><b>abc</b></td> <td><b>ijk</b></td> <td><b>xyz</b></td> </tr> </tbody> </table>