twnvy/edge-lm

twnvy/ edge-lm:latest

21 Downloads Updated 5 months ago

A token-efficient thinking model that rivals the CoT of much larger models with only 3 billion parameters.

thinking

ollama run twnvy/edge-lm

curl http://localhost:11434/api/chat \
  -d '{
    "model": "twnvy/edge-lm",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='twnvy/edge-lm',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'twnvy/edge-lm',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 5 months ago

5 months ago

eda9d7f62805 · 2.0GB ·

model

archllama

parameters3.21B

quantizationQ4_K_M

2.0GB

template

{{- if .System }}{{ .System }}{{ end }} {{- range $i, $_ := .Messages }} {{- $last := eq (len (slice

542B

system

You are an AI assistant named Edge-LM designed for inference on edge-case (or limited processing pow

425B

license

1.1kB

params

{ "num_ctx": 4096, "stop": [ "<|eot_id|>", "<|start_header_id|>" ],

100B

Readme

NOTE: This is a research preview model. Responses may not be perfect. Edge-LM is a token efficient thinking SLM that rivals the CoT and alignment of much larger models with only 3 billion parameters.

Edge-LM excels in math and data analysis. It also has a much lower hallucination rate than models of similar size, such as Llama3.2 and Phi-3.5 Mini. * Edge-LM’s score calculated based on 500 random samples due to limited compute.

Quick Start

Run Edge-LM with Ollama:

ollama run twnvy/edge-lm