U

Toppy M 7B

Undi95
4K
Context
Free
Input /1M
Free
Output /1M
N/A
Max Output
Run API Call Open in Chat

Toppy M 7B

Model Overview

Field Value
Model Name Toppy M 7B
Model ID undi95/toppy-m-7b
Author/Creator undi95
Created November 10, 2023
Parameters 7 Billion
Context Length 4,096 tokens
Architecture Mistral-based
License CC-BY-NC-4.0 (Creative Commons Attribution-NonCommercial 4.0)

Description

Toppy M 7B is a wild 7B parameter model that merges several models using the new task_arithmetic merge method from mergekit. This model combines multiple fine-tuned Mistral variants to create a versatile, uncensored text generation model.

Technical Specifications

Specification Value
Model Group Mistral
Instruct Type Alpaca
Input Modalities Text
Output Modalities Text
Default Stop Sequences ###, </s>
Format Safetensors
Framework Transformers, text-generation-inference

Supported Parameters

Parameter Description Recommended Range
temperature Controls randomness 0.7 - 1.0
top_p Nucleus sampling threshold 0.9 - 0.95
top_k Top-k sampling 40 - 100
max_tokens Maximum output length Up to 4096
repetition_penalty Reduces repetition 1.1 - 1.2
stop Stop sequences ["###", "</s>"]
Model Description Parameters
mistralai/mistral-7b-instruct Base Mistral instruction-tuned model 7B
nousresearch/nous-capybara-7b Conversational Nous Research model 7B
huggingfaceh4/zephyr-7b-beta HuggingFace optimized assistant 7B
undi95/remm-slerp-l2-13b Another merge by undi95 13B

Merged Source Models

The model is created by merging the following five models:

  1. NousResearch/Nous-Capybara-7B-V1.9 - A conversational model from Nous Research
  2. HuggingFaceH4/zephyr-7b-beta - Hugging Face's Zephyr model optimized for helpful responses
  3. lemonilia/AshhLimaRP-Mistral-7B - A roleplay-focused Mistral fine-tune
  4. Vulkane/120-Days-of-Sodom-LoRA-Mistral-7b - A LoRA adapter for creative writing
  5. Undi95/Mistral-pippa-sharegpt-7b-qlora - A QLoRA fine-tune on ShareGPT conversations

Tags

  • #merge - Created via model merging techniques
  • #uncensored - Less restrictive output filtering
  • #not-for-all-audiences - Contains mature content capabilities

Pricing

Pricing Tier Input Cost Output Cost
Free Tier $0.00 / 1M tokens $0.00 / 1M tokens
Standard Varies by provider Varies by provider

Note: Pricing may vary based on the specific provider routing through LangMart.

Prompt Format (Alpaca)

The model uses the Alpaca instruction format:

### Instruction:
{your instruction here}

### Input:
{optional input context}

### Response:
{model generates response here}

For simpler prompts without input:

### Instruction:
{your instruction here}

### Response:

Usage Example

LangMart API

curl https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "undi95/toppy-m-7b",
    "messages": [
      {
        "role": "user",
        "content": "Write a short story about a robot learning to paint."
      }
    ],
    "temperature": 0.8,
    "max_tokens": 1024
  }'

Python (OpenAI SDK Compatible)

from openai import OpenAI

client = OpenAI(
    base_url="https://api.langmart.ai/v1",
    api_key="your-langmart-api-key"
)

response = client.chat.completions.create(
    model="undi95/toppy-m-7b",
    messages=[
        {"role": "user", "content": "Explain quantum computing in simple terms."}
    ],
    temperature=0.7,
    max_tokens=512
)

print(response.choices[0].message.content)

LangMart Gateway

curl -X POST https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "undi95/toppy-m-7b",
    "messages": [
      {"role": "user", "content": "Hello, how are you?"}
    ]
  }'

Performance Notes

  • Best suited for creative writing, roleplay, and general conversation
  • The merge combines conversational, roleplay, and creative writing capabilities
  • May produce uncensored/unfiltered responses
  • Context window limited to 4,096 tokens

Content Warning

This model is tagged as "Not-For-All-Audiences" on Hugging Face, indicating it may generate mature or sensitive content. Use appropriate content filtering in production environments.

Resources

Last updated: December 2024 Data sourced from LangMart and Hugging Face