Instructions to use bobber/lex-interviewer-nemotron-4b-grpo-v12 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use bobber/lex-interviewer-nemotron-4b-grpo-v12 with Transformers:

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

pipe = pipeline("text-generation", model="bobber/lex-interviewer-nemotron-4b-grpo-v12", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("bobber/lex-interviewer-nemotron-4b-grpo-v12", trust_remote_code=True)
model = AutoModelForMultimodalLM.from_pretrained("bobber/lex-interviewer-nemotron-4b-grpo-v12", trust_remote_code=True)
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 bobber/lex-interviewer-nemotron-4b-grpo-v12 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "bobber/lex-interviewer-nemotron-4b-grpo-v12"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bobber/lex-interviewer-nemotron-4b-grpo-v12",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/bobber/lex-interviewer-nemotron-4b-grpo-v12

SGLang

How to use bobber/lex-interviewer-nemotron-4b-grpo-v12 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 "bobber/lex-interviewer-nemotron-4b-grpo-v12" \
    --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": "bobber/lex-interviewer-nemotron-4b-grpo-v12",
		"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 "bobber/lex-interviewer-nemotron-4b-grpo-v12" \
        --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": "bobber/lex-interviewer-nemotron-4b-grpo-v12",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use bobber/lex-interviewer-nemotron-4b-grpo-v12 with Docker Model Runner:
```
docker model run hf.co/bobber/lex-interviewer-nemotron-4b-grpo-v12
```

lex-interviewer-nemotron-4b-grpo-v12 / adapter /README.md

bobber

GRPO v12 LoRA adapter: score=0.760, uses_guest=60%, probing=96%

aff278c verified 3 months ago

867 Bytes

license: apache-2.0
base_model: nvidia/Nemotron-3-4B-Base
library_name: peft
pipeline_tag: text-generation
tags:
  - grpo
  - lora
  - transformers
  - trl
  - interview
  - lex-fridman

Lex Fridman Interviewer — GRPO v12 LoRA Adapter

Fine-tuned Nemotron-3-Nano-4B to ask Lex Fridman-style interview questions.

Training

LoRA v1 (r=64, LR=2e-4, 1 epoch, 4,772 pairs) → score 0.733
GRPO v12 (reward_v12, 200 steps from LoRA v1) → score 0.760

Eval (functional judge: on_topic × uses_guest × probing)

Model	Score	uses_guest	probing
Base	0.653	48%	84%
LoRA v1	0.733	56%	92%
This model	0.760	60%	96%

Reward (reward_v12)

uses_guest_logit^0.67 × probing_logit^0.33 + lexical_bonus — Qwen3.5-4B judges, continuous logits.