Instructions to use ashishnair/Llama-Ione-8B-roleplay-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ashishnair/Llama-Ione-8B-roleplay-v1 with Transformers:

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

pipe = pipeline("text-generation", model="ashishnair/Llama-Ione-8B-roleplay-v1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ashishnair/Llama-Ione-8B-roleplay-v1")
model = AutoModelForCausalLM.from_pretrained("ashishnair/Llama-Ione-8B-roleplay-v1")
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]:]))

Inference
Local Apps Settings

vLLM

How to use ashishnair/Llama-Ione-8B-roleplay-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ashishnair/Llama-Ione-8B-roleplay-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": "ashishnair/Llama-Ione-8B-roleplay-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ashishnair/Llama-Ione-8B-roleplay-v1

SGLang

How to use ashishnair/Llama-Ione-8B-roleplay-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 "ashishnair/Llama-Ione-8B-roleplay-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": "ashishnair/Llama-Ione-8B-roleplay-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 "ashishnair/Llama-Ione-8B-roleplay-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": "ashishnair/Llama-Ione-8B-roleplay-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ashishnair/Llama-Ione-8B-roleplay-v1 with Docker Model Runner:
```
docker model run hf.co/ashishnair/Llama-Ione-8B-roleplay-v1
```

ashishnair commited on Apr 9

Commit

06a22fe

verified ·

1 Parent(s): f34c3a0

Update README.md

Browse files

Files changed (1) hide show

README.md +0 -5

README.md CHANGED Viewed

@@ -126,26 +126,21 @@ The `moral_scenarios` drop is the most significant. MMLU moral scenarios test ri
 ### Stage 1 — Base merge (DARE-TIES)
 Merged `meta-llama/Llama-3.1-8B` (weight 0.3 / density 0.5) with `Gurubot/self-after-dark` (weight 0.7 / density 0.8). Personality-dominant — the character model leads; the base provides structural stability.
-Output: `merged_weird1`
 ### Stage 2 — SFT round 1
 Fine-tuned on 2,000-sample curated human texting corpus — short, emotionally varied conversational fragments.
 Train loss: 1.7368 · Runtime: 44.8 min · 2.23 samples/sec
-Output: `merged_weird1_sft`
 ### Stage 3 — Second merge (DARE-TIES)
 Merged `merged_weird1_sft` (weight 0.7 / density 0.8) with `meta-llama/Llama-3.1-8B-Instruct` (weight 0.3 / density 0.5). Recovers instruction-following capacity at low weight without overwriting persona.
-Output: `merged_B_weird2`
 ### Stage 4 — SFT round 2
 Fine-tuned on ~900-sample GPT-generated multi-persona instruction dataset covering 10+ named personas (Zoe, Maya, Iris, Hana, Tess, Kim, Vera, Cass, Nora, and others) across varied professional backgrounds and emotional states.
 Train loss: 1.1821 · Runtime: 31.4 min · 1.44 samples/sec
-Output: `merged_weird2_sft1`
 ### Stage 5 — SFT round 3 (final)
 Final fine-tuning pass on the original dialogue corpus to re-anchor conversational naturalness after instruction alignment.
 Train loss: 1.4733 · Runtime: 44.8 min · 2.23 samples/sec
-Output: **Llama-Ione-8B-roleplay-v1**
 ### Training statistics

 ### Stage 1 — Base merge (DARE-TIES)
 Merged `meta-llama/Llama-3.1-8B` (weight 0.3 / density 0.5) with `Gurubot/self-after-dark` (weight 0.7 / density 0.8). Personality-dominant — the character model leads; the base provides structural stability.
 ### Stage 2 — SFT round 1
 Fine-tuned on 2,000-sample curated human texting corpus — short, emotionally varied conversational fragments.
 Train loss: 1.7368 · Runtime: 44.8 min · 2.23 samples/sec
 ### Stage 3 — Second merge (DARE-TIES)
 Merged `merged_weird1_sft` (weight 0.7 / density 0.8) with `meta-llama/Llama-3.1-8B-Instruct` (weight 0.3 / density 0.5). Recovers instruction-following capacity at low weight without overwriting persona.
 ### Stage 4 — SFT round 2
 Fine-tuned on ~900-sample GPT-generated multi-persona instruction dataset covering 10+ named personas (Zoe, Maya, Iris, Hana, Tess, Kim, Vera, Cass, Nora, and others) across varied professional backgrounds and emotional states.
 Train loss: 1.1821 · Runtime: 31.4 min · 1.44 samples/sec
 ### Stage 5 — SFT round 3 (final)
 Final fine-tuning pass on the original dialogue corpus to re-anchor conversational naturalness after instruction alignment.
 Train loss: 1.4733 · Runtime: 44.8 min · 2.23 samples/sec
 ### Training statistics