[Examples] Add Qwen3.5-27B NVFP4A16 and MXFP4A16 quantization examples

Signed-off-by: Ziming <frankziming26@outlook.com>

Author: 2imi9

examples/quantization_w4a16_fp4/mxfp4/qwen3.5_example.py

@@ -0,0 +1,45 @@
+from compressed_tensors.offload import dispatch_model
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+from llmcompressor import oneshot
+from llmcompressor.modifiers.quantization import QuantizationModifier
+
+# Load model.
+MODEL_ID = "Qwen/Qwen3.5-27B"
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID, dtype="auto", trust_remote_code=True
+)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
+
+# Configure the quantization algorithm and scheme.
+# In this case, we:
+#   * quantize the weights to fp4 with per group 32 via ptq
+#   * skip the visual encoder, lm_head, linear attention (Gated DeltaNet
+#     fused projections are incompatible with microscale formats), and MTP modules
+recipe = QuantizationModifier(
+    targets="Linear",
+    scheme="MXFP4A16",
+    ignore=[
+        "lm_head",
+        "re:.*visual.*",
+        "re:.*linear_attn.*",
+        "re:.*mtp.*",
+    ],
+)
+
+# Apply quantization.
+oneshot(model=model, recipe=recipe)
+
+print("\n\n========== 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], skip_special_tokens=True))
+print("==========================================\n\n")
+
+# Save to disk in compressed-tensors format.
+SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-MXFP4A16"
+model.save_pretrained(SAVE_DIR, save_compressed=True)
+tokenizer.save_pretrained(SAVE_DIR)

examples/quantization_w4a16_fp4/nvfp4/qwen3.5_example.py

@@ -0,0 +1,45 @@
+from compressed_tensors.offload import dispatch_model
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+from llmcompressor import oneshot
+from llmcompressor.modifiers.quantization import QuantizationModifier
+
+# Load model.
+MODEL_ID = "Qwen/Qwen3.5-27B"
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID, dtype="auto", trust_remote_code=True
+)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
+
+# Configure the quantization algorithm and scheme.
+# In this case, we:
+#   * quantize the weights to fp4 with per group 16 via ptq
+#   * skip the visual encoder, lm_head, linear attention (Gated DeltaNet
+#     fused projections are incompatible with NVFP4), and MTP modules
+recipe = QuantizationModifier(
+    targets="Linear",
+    scheme="NVFP4A16",
+    ignore=[
+        "lm_head",
+        "re:.*visual.*",
+        "re:.*linear_attn.*",
+        "re:.*mtp.*",
+    ],
+)
+
+# Apply quantization.
+oneshot(model=model, recipe=recipe)
+
+print("\n\n========== 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], skip_special_tokens=True))
+print("==========================================\n\n")
+
+# Save to disk in compressed-tensors format.
+SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-NVFP4A16"
+model.save_pretrained(SAVE_DIR, save_compressed=True)
+tokenizer.save_pretrained(SAVE_DIR)