jsantillana/vectrayx-base-260m

jsantillana/ vectrayx-base-260m:latest

4 Downloads Updated 1 month ago

VectraYX-Base-260M is a 260M parameter LLM specialized in cybersecurity for Latin America, trained from scratch in Spanish. Supports native tool use with `<|tool_call|>`, explicit reasoning with `<think>`, and technical conversation in Latin American Span

tools

ollama run jsantillana/vectrayx-base-260m

curl http://localhost:11434/api/chat \
  -d '{
    "model": "jsantillana/vectrayx-base-260m",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='jsantillana/vectrayx-base-260m',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'jsantillana/vectrayx-base-260m',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 1 month ago

1 month ago

a36149ddd6b1 · 554MB ·

model

archllama

parameters277M

quantizationF16

554MB

system

Eres VectraYX, un asistente de ciberseguridad en español. Tienes 6 herramientas MCP disponibles: -

692B

template

{{ if .System }}<|system|>{{ .System }}<|end|>{{ end }}{{ range .Messages }}{{ if eq .Role "user" }}

298B

params

{ "num_ctx": 4096, "repeat_penalty": 1.1, "stop": [ "<|end|>", "<|user|>

152B

Readme

VectraYX-Base-260M is a 260M parameter LLM specialized in cybersecurity for Latin America, trained from scratch in Spanish. Supports native tool use with <|tool_call|>, explicit reasoning with <think>, and technical conversation in Latin American Spanish.

Capabilities: - CVE lookup (NVD/CISA KEV), IOC reputation (OTX), bash command execution - Pentesting commands: nmap, hydra, gobuster, SUID find - Chain-of-thought with <think>...</think> on complex queries - MITRE ATT&CK, incident response, threat analysis in Spanish

Available tools: nvd_get_cve · nvd_search · cisa_kev_check · otx_check_ioc · bash_exec

Architecture: 16 layers · 16 heads · GQA 4 KV · d_model=1024 · ctx=1024 · GGUF f16

Trained on 6.27B tokens (~1.2× Chinchilla optimal) across three phases: general ES pretraining → cybersecurity specialization → SFT with tool use + thinking curriculum.