Instructions to use BRR1154/SmolLM2-135M-Coder-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use BRR1154/SmolLM2-135M-Coder-v1 with Transformers:

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

pipe = pipeline("text-generation", model="BRR1154/SmolLM2-135M-Coder-v1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("BRR1154/SmolLM2-135M-Coder-v1")
model = AutoModelForCausalLM.from_pretrained("BRR1154/SmolLM2-135M-Coder-v1", 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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use BRR1154/SmolLM2-135M-Coder-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "BRR1154/SmolLM2-135M-Coder-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "BRR1154/SmolLM2-135M-Coder-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/BRR1154/SmolLM2-135M-Coder-v1

SGLang

How to use BRR1154/SmolLM2-135M-Coder-v1 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 "BRR1154/SmolLM2-135M-Coder-v1" \
    --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": "BRR1154/SmolLM2-135M-Coder-v1",
		"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 "BRR1154/SmolLM2-135M-Coder-v1" \
        --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": "BRR1154/SmolLM2-135M-Coder-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use BRR1154/SmolLM2-135M-Coder-v1 with Docker Model Runner:
```
docker model run hf.co/BRR1154/SmolLM2-135M-Coder-v1
```

SmolLM2-135M-Coder-v1: Ultra-Compact Low-Latency Edge Coding Specialist

Lead Architect: Bhavanam Rajendra Reddy (Runtime-Slayers | ORCID: 0009-0005-3349-1162)
Organization: Runtime-Slayers
Parameter Footprint: 134,516,736 (~135 Million Parameters)
Hardware & Memory Footprint: ~270 MB (FP16) / ~135 MB (INT8 / GGUF)
Base Architecture: SmolLM2-135M (Llama-style architecture, 30 Layers, 576 Hidden Dimension, 3 Key-Value Heads, 8k Context Window)
Surgical Weight Surgery: Subspace Null-Space Rank-1 Model Editing on Down-Projection MLPs ($\Delta W \cdot V_{\text{base}} = 0$)

1. Overview & Capabilities

SmolLM2-135M-Coder-v1 is an ultra-lightweight, edge-deployable Small Language Model (SLM) optimized for near-zero latency code generation, syntax validation, and algorithmic matrix calculations. Designed to run on microcontrollers, embedded robotics, and edge gateways (Raspberry Pi, Jetson Nano, WebAssembly).

Key Features:

Sub-50ms Inference Latency: Runs at >120 tokens/second on edge CPU.
Micro-Memory Footprint: Under 300MB RAM requirement.
Syntax Grounding: Specially edited for clean Python, MATLAB, and G-Code routines.

2. Quickstart Inference Example

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "BRR1154/SmolLM2-135M-Coder-v1"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float32,
    device_map="auto"
)

prompt = "Write a vectorized Python function using NumPy to compute the Euclidean distance matrix between two sets of points."
messages = [{"role": "user", "content": prompt}]
inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(model.device)

outputs = model.generate(inputs, max_new_tokens=256, temperature=0.1)
print(tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True))

📊 Empirical Academic Benchmarks & Evaluation Realities

1. Measured Academic Benchmarks (`lm-eval` v0.4.13 on CPU):

Benchmark	Task Focus	Sample Size	Raw Accuracy	Normalized (`acc_norm`)	Standard Error	Real-World Takeaway
ARC-Easy	Elementary grade science QA	50	50.0% (25/50)	44.0%	$\pm 7.14%$	Moderate baseline matching base SmolLM2.
ARC-Challenge	Multi-hop complex scientific deduction	50	26.0% (13/50)	26.0%	$\pm 6.27%$	Near random 4-choice guess baseline (25%).
HellaSwag	Situational continuation	50	44.0% (22/50)	44.0%	$\pm 7.09%$	Typical for ultra-compact 135M architectures.
OpenBookQA	Multi-hop open-book facts	50	16.0% (8/50)	38.0%	$\pm 6.93%$	Below random baseline on raw accuracy.
WinoGrande	Coreference ambiguity resolution	50	52.0% (26/50)	52.0%	$\pm 7.14%$	Near 50% binary choice baseline.

2. Official Foundation Baselines (`SmolLM2-135M-Instruct` Technical Report, 2024):

MMLU (5-shot): 25.8% (barely above 25% random baseline)
GSM8K (8-shot Math): 8.3% (arithmetic reasoning requires external tools)
HumanEval (Pass@1, Python): 22.0%

3. Real Codebase Comprehension & Ambiguity Limitations:

Repository Architecture Comprehension: 42.17 / 100 (Tested on real scientific codebases new_planet_identification and Bio-Sync-AI).
Ambiguity & Clarification Stress Test: 0.0 / 8.0 (0%). When presented with ambiguous deployment requests without credentials, the 135M parameter capacity defaults to emitting code completions rather than recognizing missing information and asking for clarification.
Architectural Role: This model functions as an ultra-compact, low-latency, deterministic syntax and token emitter (sub-50ms inference on edge CPU). High-level cognitive arbitration and multi-hop reasoning are delegated to larger models or closed-loop execution harnesses.

3. Citation

@misc{reddy2026smollmcoder,
  author = {Reddy, Bhavanam Rajendra},
  title = {SmolLM2-135M-Coder-v1: Ultra-Compact Edge Coding Specialist},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\\url{https://huggingface.co/BRR1154/SmolLM2-135M-Coder-v1}}
}

Downloads last month: -

Safetensors

Model size

0.1B params

Tensor type

F32

Model tree for BRR1154/SmolLM2-135M-Coder-v1

Base model

HuggingFaceTB/SmolLM2-135M

Quantized

HuggingFaceTB/SmolLM2-135M-Instruct

Finetuned

(411)

this model