Instructions to use SohamProgrammer/Unrealistic-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SohamProgrammer/Unrealistic-v1 with Transformers:

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

pipe = pipeline("text-generation", model="SohamProgrammer/Unrealistic-v1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("SohamProgrammer/Unrealistic-v1")
model = AutoModelForCausalLM.from_pretrained("SohamProgrammer/Unrealistic-v1", device_map="auto")

MLX

How to use SohamProgrammer/Unrealistic-v1 with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("SohamProgrammer/Unrealistic-v1")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use SohamProgrammer/Unrealistic-v1 with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf SohamProgrammer/Unrealistic-v1:F16
# Run inference directly in the terminal:
llama cli -hf SohamProgrammer/Unrealistic-v1:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf SohamProgrammer/Unrealistic-v1:F16
# Run inference directly in the terminal:
llama cli -hf SohamProgrammer/Unrealistic-v1:F16

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf SohamProgrammer/Unrealistic-v1:F16
# Run inference directly in the terminal:
./llama-cli -hf SohamProgrammer/Unrealistic-v1:F16

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf SohamProgrammer/Unrealistic-v1:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf SohamProgrammer/Unrealistic-v1:F16

Use Docker

docker model run hf.co/SohamProgrammer/Unrealistic-v1:F16

LM Studio
Jan

vLLM

How to use SohamProgrammer/Unrealistic-v1 with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/SohamProgrammer/Unrealistic-v1:F16

SGLang

How to use SohamProgrammer/Unrealistic-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 "SohamProgrammer/Unrealistic-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": "SohamProgrammer/Unrealistic-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 "SohamProgrammer/Unrealistic-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": "SohamProgrammer/Unrealistic-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use SohamProgrammer/Unrealistic-v1 with Ollama:
```
ollama run hf.co/SohamProgrammer/Unrealistic-v1:F16
```
Unsloth Desktop

MLX LM

How to use SohamProgrammer/Unrealistic-v1 with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "SohamProgrammer/Unrealistic-v1"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "SohamProgrammer/Unrealistic-v1"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "SohamProgrammer/Unrealistic-v1",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Docker Model Runner
How to use SohamProgrammer/Unrealistic-v1 with Docker Model Runner:
```
docker model run hf.co/SohamProgrammer/Unrealistic-v1:F16
```

Lemonade

How to use SohamProgrammer/Unrealistic-v1 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull SohamProgrammer/Unrealistic-v1:F16

Run and chat with the model

lemonade run user.Unrealistic-v1-F16

List all available models

lemonade list

Atomic Chat

Unrealistic v1 (190M)

Decoder-only transformer LM (189,748,992 params) trained from scratch on a MacBook Air M4 with MLX. LLaMA-family: 14 layers, hidden 768, FFN 4096, 8 heads, RoPE θ=10k, RMSNorm, SiLU/SwiGLU, tied embeddings, bf16, 32K SentencePiece vocab, 1024 context.

Training

~356K steps / ~1.1B tokens. Chain: Stage C → CoT (math+code reasoning) → Refresh (240M topper) → Realization (judgment tasks) → SFT (31.7M incl. OpenAssistant) → Wiki entities+math (13.35M) → SFT-refresh → DPO (UltraFeedback, β=0.5) → diverse-SFT (smol-smoltalk + UltraChat + task synthetics) → pristine anchors. AdamW throughout. Full log: docs/TRAINING_REPORT.md.

Evaluation (final `step_355789`)

Math free-gen 10/15 · GK 16/16 · Code 3/5 · Culture 4/7
First-token probes 20/40 top-3
Standard benchmarks: evals/benchmarks/final/results.json (ARC-Easy, PIQA, HellaSwag-sample, Winogrande, TruthfulQA-MC1, MMLU-easy, GSM8K-sample, MBPP-sample)
Full detail: docs/EVAL_REPORT.md

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("REPO_ID")
model = AutoModelForCausalLM.from_pretrained("REPO_ID")

On Apple Silicon the native MLX checkpoint + scripts/spot_infer.py is fastest. Suggested decode: temperature 0.4, top_p 0.9, repetition_penalty 1.25.

Raw GGUF (unrealistic-v1-f16.gguf, exact weights, no quantization) + a local-Ollama Modelfile ship alongside. (Ollama library listing excluded: publisher under 18; local use unaffected.)

Ollama (recommended setup — required for correct behavior)

Direct ollama run hf.co/... uses wrong sampling defaults (temp 0.8) and may ignore the embedded template. One-command correct setup (fetches the tested Modelfile, builds the record — no manual config):

curl -sL https://huggingface.co/SohamProgrammer/Unrealistic-v1/resolve/main/setup-ollama.sh | bash
ollama run unrealistic-v1

(Manual alternative: download Modelfile from Files, ollama create unrealistic-v1 -f Modelfile.) Note: FROM hf.co/... Modelfiles fail on this repo because HF serves large files via Xet CDN, whose cross-host redirects Ollama blocks. The installer above downloads first and creates locally — the supported path.

This pins temperature 0.4, top_p 0.9, repeat_penalty 1.25, ctx 1024, the User:/Assistant: chat template, and User: stop — the exact configuration the model was validated with. No system prompt (untrained distribution). Phone users: re-download the Q4 file (post-fix bytes with BOS disabled).

Limitations

Weak: large-number word problems, deep code, biology, multi-step reasoning, instruction-following beyond basics, non-English. May hallucinate entities and dates. 190M memorization model, not a reasoner.

Data & license

Weights: Apache-2.0. Training-data attribution: docs/DATA_REPORT.md (CC BY-SA content attributed; 0.6% GPL-family trace in code data measured and documented; no verbatim reproduction possible at this scale).

Downloads last month: 1,307

Safetensors

Model size

0.2B params

Tensor type

BF16

MLX

Hardware compatibility

Quantized

SohamProgrammer
/

Unrealistic-v1

Unrealistic v1 (190M)

Training

Evaluation (final `step_355789`)

Usage

Ollama (recommended setup — required for correct behavior)

Limitations

Data & license

Space using SohamProgrammer/Unrealistic-v1 1

Unrealistic v1 (190M)

Training

Evaluation (final step_355789)

Usage

Ollama (recommended setup — required for correct behavior)

Limitations

Data & license

Space using SohamProgrammer/Unrealistic-v1 1

Evaluation (final `step_355789`)