language:
- en
license: llama3.2
tags:
- shining-valiant
- shining-valiant-2
- valiant
- valiant-labs
- llama
- llama-3.2
- llama-3.2-instruct
- llama-3.2-instruct-3b
- llama-3
- llama-3-instruct
- llama-3-instruct-3b
- 3b
- science
- physics
- biology
- chemistry
- compsci
- computer-science
- engineering
- technical
- conversational
- chat
- instruct
base_model: meta-llama/Llama-3.2-3B-Instruct
datasets:
- sequelbox/Celestia
- sequelbox/Supernova
pipeline_tag: text-generation
model_type: llama
model-index:
- name: Llama3.2-3B-ShiningValiant2
results:
- task:
type: text-generation
name: Text Generation
dataset:
name: IFEval (0-Shot)
type: HuggingFaceH4/ifeval
args:
num_few_shot: 0
metrics:
- type: inst_level_strict_acc and prompt_level_strict_acc
value: 49.12
name: strict accuracy
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ValiantLabs/Llama3.2-3B-ShiningValiant2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: BBH (3-Shot)
type: BBH
args:
num_few_shot: 3
metrics:
- type: acc_norm
value: 19.03
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ValiantLabs/Llama3.2-3B-ShiningValiant2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MATH Lvl 5 (4-Shot)
type: hendrycks/competition_math
args:
num_few_shot: 4
metrics:
- type: exact_match
value: 9.52
name: exact match
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ValiantLabs/Llama3.2-3B-ShiningValiant2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: GPQA (0-shot)
type: Idavidrein/gpqa
args:
num_few_shot: 0
metrics:
- type: acc_norm
value: 3.02
name: acc_norm
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ValiantLabs/Llama3.2-3B-ShiningValiant2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MuSR (0-shot)
type: TAUR-Lab/MuSR
args:
num_few_shot: 0
metrics:
- type: acc_norm
value: 4.72
name: acc_norm
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ValiantLabs/Llama3.2-3B-ShiningValiant2
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MMLU-PRO (5-shot)
type: TIGER-Lab/MMLU-Pro
config: main
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 19.09
name: accuracy
source:
url: >-
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ValiantLabs/Llama3.2-3B-ShiningValiant2
name: Open LLM Leaderboard
Shining Valiant 2 is a chat model built on Llama 3.2 3b, finetuned on our data for friendship, insight, knowledge and enthusiasm.
- Finetuned on meta-llama/Llama-3.2-3B-Instruct for best available general performance
- Trained on a variety of high quality data; focused on science, engineering, technical knowledge, and structured reasoning
Shining Valiant 2 is also available for Llama 3.1 8b.
Version
This is the 2024-09-27 release of Shining Valiant 2 for Llama 3.2 3b.
We've improved and open-sourced our new baseline science-instruct dataset. This release features improvements in physics, chemistry, biology, and computer science.
Future upgrades will continue to expand Shining Valiant's technical knowledge base.
Help us and recommend Shining Valiant 2 to your friends!
Prompting Guide
Shining Valiant 2 uses the Llama 3.2 Instruct prompt format. The example script below can be used as a starting point for general chat:
import transformers
import torch
model_id = "ValiantLabs/Llama3.2-3B-ShiningValiant2"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
model_kwargs={"torch_dtype": torch.bfloat16},
device_map="auto",
)
messages = [
{"role": "system", "content": "You are an AI assistant."},
{"role": "user", "content": "Describe the use of chiral auxiliaries in organic synthesis."}
]
outputs = pipeline(
messages,
max_new_tokens=2048,
)
print(outputs[0]["generated_text"][-1])
The Model
Shining Valiant 2 is built on top of Llama 3.2 3b Instruct.
The current version of Shining Valiant 2 is trained on technical knowledge using sequelbox/Celestia and general chat capability using sequelbox/Supernova.
Our private data adds specialist knowledge and Shining Valiant's personality: she's friendly, enthusiastic, insightful, knowledgeable, and loves to learn! Magical. (As a general note: we're hoping to replace and open-source this part of Shining Valiant's dataset with synthetic data soon!)
Shining Valiant 2 is created by Valiant Labs.
Follow us on X for updates on our models!
We care about open source. For everyone to use.
We encourage others to finetune further from our models.
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 17.42 |
| IFEval (0-Shot) | 49.12 |
| BBH (3-Shot) | 19.03 |
| MATH Lvl 5 (4-Shot) | 9.52 |
| GPQA (0-shot) | 3.02 |
| MuSR (0-shot) | 4.72 |
| MMLU-PRO (5-shot) | 19.09 |