lil-c3po / README.md

Adding Evaluation Results (#3)

5954ab3 verified 3 months ago

No virus

5.23 kB

	---
	language:
	- en
	license: mit
	model-index:
	- name: lil-c3po
	results:
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: AI2 Reasoning Challenge (25-Shot)
	type: ai2_arc
	config: ARC-Challenge
	split: test
	args:
	num_few_shot: 25
	metrics:
	- type: acc_norm
	value: 65.02
	name: normalized accuracy
	source:
	url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=deepnight-research/lil-c3po
	name: Open LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: HellaSwag (10-Shot)
	type: hellaswag
	split: validation
	args:
	num_few_shot: 10
	metrics:
	- type: acc_norm
	value: 84.45
	name: normalized accuracy
	source:
	url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=deepnight-research/lil-c3po
	name: Open LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: MMLU (5-Shot)
	type: cais/mmlu
	config: all
	split: test
	args:
	num_few_shot: 5
	metrics:
	- type: acc
	value: 62.36
	name: accuracy
	source:
	url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=deepnight-research/lil-c3po
	name: Open LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: TruthfulQA (0-shot)
	type: truthful_qa
	config: multiple_choice
	split: validation
	args:
	num_few_shot: 0
	metrics:
	- type: mc2
	value: 68.73
	source:
	url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=deepnight-research/lil-c3po
	name: Open LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: Winogrande (5-shot)
	type: winogrande
	config: winogrande_xl
	split: validation
	args:
	num_few_shot: 5
	metrics:
	- type: acc
	value: 79.16
	name: accuracy
	source:
	url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=deepnight-research/lil-c3po
	name: Open LLM Leaderboard
	- task:
	type: text-generation
	name: Text Generation
	dataset:
	name: GSM8k (5-shot)
	type: gsm8k
	config: main
	split: test
	args:
	num_few_shot: 5
	metrics:
	- type: acc
	value: 48.45
	name: accuracy
	source:
	url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=deepnight-research/lil-c3po
	name: Open LLM Leaderboard
	---

	# deepnight-research/lil-c3po
	<div style="display: flex; justify-content: center; align-items: center;">
	<img src="./lil-c3po.jpg" style="width: 100%; height: auto;"/></div>

	## Model Details:
	lil-c3po is an open-source large language model (LLM) resulting from the linear merge of two distinct
	fine-tuned Mistral-7B models, internally referred to as c3-1 and c3-2. These models, developed in-house,
	bring together unique characteristics to enhance performance and utility.

	## Model Architecture:
	lil-c3po inherits its architecture from the combined c3-1 and c3-2 models,
	incorporating features such as Grouped-Query Attention, Sliding-Window Attention, and Byte-fallback BPE tokenizer.
	This fusion aims to capitalize on the strengths of both models for improved language understanding and generation.

	## Training Details:
	- The first model, internally referred to as c3-1, is a 7B parameter Large Language Model
	fine-tuned on the Intel Gaudi 2 processor.
	It utilizes the Direct Performance Optimization (DPO) method and is designed to excel in various language-related tasks.
	- The second model, denoted as c3-2, is an instruct fine-tuned version of Mistral-7B.
	Its architecture features improvements in instruct fine-tuning, contributing to enhanced language understanding in instructional contexts.

	## License:
	lil-c3po is released under the MIT license, fostering open-source collaboration and innovation.

	## Intended Use:
	This merged model is suitable for a broad range of language-related tasks,
	inheriting the capabilities of the fine-tuned c3-1 and c3-2 models. Users interested in language tasks can leverage lil-c3po's capabilities.

	## Out-of-Scope Uses:
	While lil-c3po is versatile, it is important to note that, in most cases, fine-tuning may be necessary for specific tasks.
	Additionally, the model should not be used to intentionally create hostile or alienating environments for people.
	# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
	Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_deepnight-research__lil-c3po)

	\| Metric \|Value\|
	\|---------------------------------\|----:\|
	\|Avg. \|68.03\|
	\|AI2 Reasoning Challenge (25-Shot)\|65.02\|
	\|HellaSwag (10-Shot) \|84.45\|
	\|MMLU (5-Shot) \|62.36\|
	\|TruthfulQA (0-shot) \|68.73\|
	\|Winogrande (5-shot) \|79.16\|
	\|GSM8k (5-shot) \|48.45\|