add fp8 block example

examples/autoround/README.md

@@ -68,6 +68,7 @@ The accuracy of the quantized model is configured by tuning-related parameters.
 | `wNa16` + `FP8KV`   | [llama3_example](./quantization_kv_cache/llama3_example.py)               |                                       |
 | `W8A8-FP8` Static   | [llama4_example](./quantization_w8a8_fp8/llama4_static_quant_example.py) |                                       |
 | `W8A8-FP8` Dynamic  | [llama4_example](./quantization_w8a8_fp8/llama4_dynamic_quant_example.py)  |                                       |
+| `W8A8-FP8` Block    | [llama3.1_example](./quantization_w8a8_fp8/llama3.1_block_quant_example.py) |                                     |
 | `NVFP4`  | [llama3.1_example](./quantization_w4a4_fp4/llama3.1_example.py)  |                                       |
 | `MXFP4`  | [qwen3_example](../../experimental/mxfp4/autoround_qwen3_example.py)  |                                       |
 

examples/autoround/quantization_w8a8_fp8/llama3.1_block_quant_example.py

@@ -0,0 +1,56 @@
+from auto_round.calib_dataset import get_dataset
+from compressed_tensors.offload import dispatch_model
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+from llmcompressor import oneshot
+from llmcompressor.modifiers.autoround import AutoRoundModifier
+
+# Select model and load it.
+MODEL_ID = "meta-llama/Meta-Llama-3.1-8B-Instruct"
+model = AutoModelForCausalLM.from_pretrained(MODEL_ID, dtype="auto")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+
+# Select calibration dataset.
+NUM_CALIBRATION_SAMPLES = 128
+MAX_SEQUENCE_LENGTH = 2048
+# Get aligned calibration dataset.
+
+ds = get_dataset(
+    tokenizer=tokenizer,
+    seqlen=MAX_SEQUENCE_LENGTH,
+    nsamples=NUM_CALIBRATION_SAMPLES,
+)
+
+
+# Configure the quantization algorithm to run.
+# NOTE: AutoRoundModifier with iters=0 is equivalent to RTN
+recipe = AutoRoundModifier(
+    targets="Linear", scheme="FP8_BLOCK", ignore=["lm_head"], iters=0
+)
+
+
+# Apply algorithms.
+oneshot(
+    model=model,
+    dataset=ds,
+    recipe=recipe,
+    max_seq_length=MAX_SEQUENCE_LENGTH,
+    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
+    shuffle_calibration_samples=False,
+)
+
+# Confirm generations of the quantized model look sane.
+print("\n\n")
+print("========== SAMPLE GENERATION ==============")
+dispatch_model(model)
+input_ids = tokenizer("Hello my name is", return_tensors="pt").input_ids.to(
+    model.device
+)
+output = model.generate(input_ids, max_new_tokens=100)
+print(tokenizer.decode(output[0]))
+print("==========================================\n\n")
+
+# Save to disk compressed.
+SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-FP8-BLOCK-AutoRound"
+model.save_pretrained(SAVE_DIR, save_compressed=True)
+tokenizer.save_pretrained(SAVE_DIR)