Instructions to use rodrigoramosrs/veriloop-coder-e2-nvfp4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use rodrigoramosrs/veriloop-coder-e2-nvfp4 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="rodrigoramosrs/veriloop-coder-e2-nvfp4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("rodrigoramosrs/veriloop-coder-e2-nvfp4")
model = AutoModelForCausalLM.from_pretrained("rodrigoramosrs/veriloop-coder-e2-nvfp4", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use rodrigoramosrs/veriloop-coder-e2-nvfp4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "rodrigoramosrs/veriloop-coder-e2-nvfp4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "rodrigoramosrs/veriloop-coder-e2-nvfp4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/rodrigoramosrs/veriloop-coder-e2-nvfp4

SGLang

How to use rodrigoramosrs/veriloop-coder-e2-nvfp4 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "rodrigoramosrs/veriloop-coder-e2-nvfp4" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "rodrigoramosrs/veriloop-coder-e2-nvfp4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "rodrigoramosrs/veriloop-coder-e2-nvfp4" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "rodrigoramosrs/veriloop-coder-e2-nvfp4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use rodrigoramosrs/veriloop-coder-e2-nvfp4 with Docker Model Runner:
```
docker model run hf.co/rodrigoramosrs/veriloop-coder-e2-nvfp4
```

rodrigoramosrs commited on 2 days ago

Commit

252883c

verified ·

1 Parent(s): 907eff1

Upload scripts/quantize_veriloop.py with huggingface_hub

Browse files

Files changed (1) hide show

scripts/quantize_veriloop.py +19 -10

scripts/quantize_veriloop.py CHANGED Viewed

@@ -1,8 +1,10 @@
 """NVFP4 W4A4 quantization for VeriLoop-E2 (Qwen3.8-27B based).
-Uses NVIDIA Model Optimizer's canonical recipe verbatim:
-    NVFP4_W4A4_WEIGHT_LOCAL_HESSIAN_CFG
-(local Hessian + fp8 scale sweep, static weight scales + dynamic input scales).
 Calibration: nvidia/Nemotron-Competitive-Programming-v1 (streaming), defaults
 512 samples x 512 tokens (262144 tokens total).
@@ -93,13 +95,20 @@ def build_device_map():
 def build_quant_cfg():
-    # NVIDIA canonical recipe, VERBATIM (no overrides):
-    # local_hessian + fp8_scale_sweep, global *weight_quantizer (static) +
-    # *input_quantizer (dynamic), with the right exclusions
-    # (mtp/visual/lm_head/embeddings). Extra overrides only duplicate
-    # calibration state; a forced layerwise mode grows an unbounded per-layer
-    # capture cache. The base global mode does not accumulate.
-    return copy.deepcopy(mtq.NVFP4_W4A4_WEIGHT_LOCAL_HESSIAN_CFG)
 def messages_to_text(messages):

 """NVFP4 W4A4 quantization for VeriLoop-E2 (Qwen3.8-27B based).
+Uses NVIDIA Model Optimizer's canonical recipe
+    (NVFP4_W4A4_WEIGHT_LOCAL_HESSIAN_CFG: local Hessian + fp8 scale sweep,
+    static weight scales + dynamic input scales) with linear_attn blocks and
+    self-attention projections kept in BF16, matching validated NVFP4
+    releases for this architecture family.
 Calibration: nvidia/Nemotron-Competitive-Programming-v1 (streaming), defaults
 512 samples x 512 tokens (262144 tokens total).
 def build_quant_cfg():
+    # NVIDIA canonical recipe + granularity adjustments: linear_attn (GDN)
+    # fully BF16 plus BF16 self-attention, matching validated NVFP4 releases
+    # for this architecture family (MLP-only NVFP4). NVFP4 attention/GDN
+    # weights produce degenerate output on some stacks; MLP-only is the
+    # widely-deployed pattern (conv1d/in_proj_a/in_proj_b already disabled
+    # in the base recipe). Appended last: entries apply in list order,
+    # later overrides earlier.
+    cfg = copy.deepcopy(mtq.NVFP4_W4A4_WEIGHT_LOCAL_HESSIAN_CFG)
+    for name in ["*linear_attn.in_proj_qkv*", "*linear_attn.in_proj_z*",
+                 "*linear_attn.out_proj*",
+                 "*self_attn.q_proj*", "*self_attn.k_proj*",
+                 "*self_attn.v_proj*", "*self_attn.o_proj*"]:
+        cfg["quant_cfg"].append({"quantizer_name": name, "enable": False})
+    return cfg
 def messages_to_text(messages):