Instructions to use VTXAI/vortex-50m with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use VTXAI/vortex-50m with Transformers:

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

pipe = pipeline("text-generation", model="VTXAI/vortex-50m", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("VTXAI/vortex-50m", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use VTXAI/vortex-50m with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "VTXAI/vortex-50m"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "VTXAI/vortex-50m",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/VTXAI/vortex-50m

SGLang

How to use VTXAI/vortex-50m 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 "VTXAI/vortex-50m" \
    --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": "VTXAI/vortex-50m",
		"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 "VTXAI/vortex-50m" \
        --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": "VTXAI/vortex-50m",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use VTXAI/vortex-50m with Docker Model Runner:
```
docker model run hf.co/VTXAI/vortex-50m
```

Abhaykoul commited on 21 days ago

Commit

cc6ed83

verified ·

1 Parent(s): 64ecbeb

Add HF configuration + modeling modules

Browse files

Files changed (2) hide show

README.md +8 -25
model.safetensors +1 -1

README.md CHANGED Viewed

@@ -12,41 +12,24 @@ tags:
 - sft
 ---
-# Vortex-50M (Cybersecurity & Cryptography Specialist)
-**Vortex-50M** is an on-device, sub-50M parameter decoder-only transformer with a 16,384-token vocabulary, fine-tuned (SFT) as a **domain-specific technical assistant for Cybersecurity, Post-Quantum Cryptography, and Security Protocols** on [`VTXAI/cyber-crypto-qa-50k`](https://huggingface.co/datasets/VTXAI/cyber-crypto-qa-50k) from the base pretrained model [`VTXAI/vortex-50m-16k`](https://huggingface.co/VTXAI/vortex-50m-16k).
 ## Model Architecture
-- **Parameters**: 49,846,528 (~49.85M, within 50M competition budget)
 - **Layers**: 18
 - **Hidden dimension**: 512
 - **Intermediate dimension**: 1072
 - **Attention heads**: 8 query heads, 2 key/value heads (Grouped Query Attention)
 - **Context length**: 2048
-- **Vocabulary size**: 16,387 (tied embeddings, with ChatML control tokens)
 - **Positional Embeddings**: RoPE (Rotary Position Embeddings)
 - **Normalization**: RMSNorm with QK-Norm
-## Benchmark Results (`eval_competition.py`)
-Evaluated with `lm-evaluation-harness` and WikiText-103:
-| Benchmark | Primary Metric | Score | Stderr |
-|:---|:---:|:---:|:---:|
-| **PIQA** | `acc_norm` | **54.79%** | ±1.16% |
-| **WinoGrande** | `acc` | **51.46%** | ±1.40% |
-| **ARC-Easy** | `acc_norm` | **33.63%** | ±0.97% |
-| **HellaSwag** | `acc_norm` | **27.10%** | ±0.44% |
-| **Scored Average (4 tasks)** | - | **41.74%** | - |
-| **WikiText-103** | Perplexity | **1394.36** | 7.24 nats (10.45 bpb) |
-## Specialized Domain Capabilities
-Trained on technical Q&A across:
-- **Network Security & Penetration Testing**: TLS/SSL handshakes, C2 beaconing detection, DDoS mitigation, authentication protocols, penetration testing methodologies.
-- **Quantum & Classical Cryptography**: BB84 QKD protocol, E91, Post-Quantum Cryptography (PQC), lattice cryptography, Harvest-Now-Decrypt-Later (HNDL), quantum bit error rates (QBER).
-- **Application Security**: Two-factor authentication (2FA), API security, risk assessment, encryption standards (AES, RSA, ECC).
-## Quickstart & Usage
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
@@ -61,6 +44,6 @@ messages = [
 prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
 inputs = tokenizer(prompt, return_tensors="pt")
-outputs = model.generate(**inputs, max_new_tokens=80, do_sample=True, temperature=0.3, repetition_penalty=1.15)
 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```

 - sft
 ---
+# Vortex-50M (Cybersecurity & Cryptography Specialist - Low Perplexity Edition)
+**Vortex-50M** is an on-device, sub-50M parameter decoder-only transformer with a 16,384-token vocabulary, fine-tuned (SFT) on [`VTXAI/cyber-crypto-balanced-qa`](https://huggingface.co/datasets/VTXAI/cyber-crypto-balanced-qa) from the base pretrained model [`VTXAI/vortex-50m-16k`](https://huggingface.co/VTXAI/vortex-50m-16k).
+This edition balances high-fidelity technical Cybersecurity & Cryptography QA with strong general language modeling regularization, preventing catastrophic forgetting and keeping perplexity low.
 ## Model Architecture
+- **Parameters**: 49,846,528 (~49.85M, within 50M budget)
 - **Layers**: 18
 - **Hidden dimension**: 512
 - **Intermediate dimension**: 1072
 - **Attention heads**: 8 query heads, 2 key/value heads (Grouped Query Attention)
 - **Context length**: 2048
+- **Vocabulary size**: 16,387 (tied embeddings)
 - **Positional Embeddings**: RoPE (Rotary Position Embeddings)
 - **Normalization**: RMSNorm with QK-Norm
+## Usage
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
 prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
 inputs = tokenizer(prompt, return_tensors="pt")
+outputs = model.generate(**inputs, max_new_tokens=60, do_sample=True, temperature=0.3, repetition_penalty=1.15)
 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```

model.safetensors CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fbd4fba1fc19deb45421bbffb0bdb05069974b3e0e2228e32065fb2061b58fcb
 size 199407088

 version https://git-lfs.github.com/spec/v1
+oid sha256:b3e540223aabbdb580f4e7e78f8fa66e479a49dc4b52cb9fd6e6d62230b86c4a
 size 199407088