| Quantization made by Richard Erkhov. | |
| [Github](https://github.com/RichardErkhov) | |
| [Discord](https://discord.gg/pvy7H8DZMG) | |
| [Request more models](https://github.com/RichardErkhov/quant_request) | |
| Memphis-CoT-3B - GGUF | |
| - Model creator: https://huggingface.co/euclaise/ | |
| - Original model: https://huggingface.co/euclaise/Memphis-CoT-3B/ | |
| | Name | Quant method | Size | | |
| | ---- | ---- | ---- | | |
| | [Memphis-CoT-3B.Q2_K.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q2_K.gguf) | Q2_K | 1.01GB | | |
| | [Memphis-CoT-3B.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.IQ3_XS.gguf) | IQ3_XS | 1.11GB | | |
| | [Memphis-CoT-3B.IQ3_S.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.IQ3_S.gguf) | IQ3_S | 1.17GB | | |
| | [Memphis-CoT-3B.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q3_K_S.gguf) | Q3_K_S | 1.17GB | | |
| | [Memphis-CoT-3B.IQ3_M.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.IQ3_M.gguf) | IQ3_M | 1.23GB | | |
| | [Memphis-CoT-3B.Q3_K.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q3_K.gguf) | Q3_K | 1.3GB | | |
| | [Memphis-CoT-3B.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q3_K_M.gguf) | Q3_K_M | 1.3GB | | |
| | [Memphis-CoT-3B.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q3_K_L.gguf) | Q3_K_L | 1.4GB | | |
| | [Memphis-CoT-3B.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.IQ4_XS.gguf) | IQ4_XS | 1.43GB | | |
| | [Memphis-CoT-3B.Q4_0.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q4_0.gguf) | Q4_0 | 1.5GB | | |
| | [Memphis-CoT-3B.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.IQ4_NL.gguf) | IQ4_NL | 1.51GB | | |
| | [Memphis-CoT-3B.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q4_K_S.gguf) | Q4_K_S | 1.51GB | | |
| | [Memphis-CoT-3B.Q4_K.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q4_K.gguf) | Q4_K | 1.59GB | | |
| | [Memphis-CoT-3B.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q4_K_M.gguf) | Q4_K_M | 1.59GB | | |
| | [Memphis-CoT-3B.Q4_1.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q4_1.gguf) | Q4_1 | 1.65GB | | |
| | [Memphis-CoT-3B.Q5_0.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q5_0.gguf) | Q5_0 | 1.81GB | | |
| | [Memphis-CoT-3B.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q5_K_S.gguf) | Q5_K_S | 1.81GB | | |
| | [Memphis-CoT-3B.Q5_K.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q5_K.gguf) | Q5_K | 1.86GB | | |
| | [Memphis-CoT-3B.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q5_K_M.gguf) | Q5_K_M | 1.86GB | | |
| | [Memphis-CoT-3B.Q5_1.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q5_1.gguf) | Q5_1 | 1.96GB | | |
| | [Memphis-CoT-3B.Q6_K.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q6_K.gguf) | Q6_K | 2.14GB | | |
| | [Memphis-CoT-3B.Q8_0.gguf](https://huggingface.co/RichardErkhov/euclaise_-_Memphis-CoT-3B-gguf/blob/main/Memphis-CoT-3B.Q8_0.gguf) | Q8_0 | 2.77GB | | |
| Original model description: | |
| --- | |
| license: cc-by-sa-3.0 | |
| library_name: transformers | |
| tags: | |
| - supertrainer2000 | |
| - human-data | |
| datasets: | |
| - euclaise/TinyCoT | |
| - euclaise/reddit-instruct | |
| - sablo/oasst2_curated | |
| - euclaise/SciCoT | |
| metrics: | |
| - accuracy | |
| base_model: stabilityai/stablelm-3b-4e1t | |
| --- | |
| *Now with a training bug fixed!* | |
| %3C!-- HTML_TAG_END --> | |
| Memphis-CoT is a finetune of [StableLM 3b 4e1t](stabilityai/stablelm-3b-4e1t) on [TinyCoT](https://huggingface.co/datasets/euclaise/TinyCoT), [SciCoT](https://huggingface.co/datasets/euclaise/SciCoT), along with [reddit-instruct](https://huggingface.co/datasets/euclaise/reddit-instruct) (subset to 5000 examples, excluding posts with brackets in the title) and a [curated](https://huggingface.co/datasets/sablo/oasst2_curated) subset of [oasst2](https://huggingface.co/datasets/OpenAssistant/oasst2). | |
| **Memphis was trained *only* on human data! No GPT generations here.** | |
| Finetuning was performed using my [supertrainer2000](https://github.com/euclaise/supertrainer2000) framework, using my Adalite optimizer. | |
| ## Training Procedure | |
| I finetuned the model using an iterative rationale-bootstrapping procedure inspired by [STaR](https://research.google/pubs/star-self-taught-reasoner-bootstrapping-reasoning-with-reasoning/) and [SPIN](https://arxiv.org/abs/2401.01335) | |
| First, I finetuned the model on all the datasets using a [MixCE](https://arxiv.org/abs/2305.16958) loss and [NEFTune](https://arxiv.org/abs/2310.05914), for 2 epochs. | |
| I then performed the following steps 3 times: | |
| 1. Generate responses for each question in TinyCoT using the current model, check each response for correctness, and create a dataset of (correct, incorrect) pairs. Extra values are discarded, such that each correct and incorrect response is unique. | |
| 2. Finetune the model for 1 epoch using a ranking loss over length-normalized log-probabilities of each sequence, similar to [Preference Ranking Optimization](https://arxiv.org/abs/2306.17492), comparing the correct vs incorrect generated response. Additionally, a standard CE loss over the chosen completion was included. | |
| This should be more efficient than either STaR or SPIN, as it uses a ranking loss rather than rejection sampling (unlike STaR), and verifies correctness instead of assuming all model responses are incorrect (unlike SPIN). | |
| To prevent excessive drift, I kept the model weights as a moving average: After each generate+train cycle, I interpolated between the previous model weights and the updated weights using spherical linear interpolation (SLERP), with an interpolation factor of 0.99. | |
| ## Prompt formats | |
| The format for reddit-instruct and oasst2 was: | |
| ``` | |
| ### User: | |
| [insert instruction here] | |
| ### Assistant: | |
| [insert response here] | |
| ### User: | |
| ... | |
| ``` | |
| The format for TinyCoT was: | |
| ``` | |
| ### User: | |
| [insert instruction here] | |
| ### Rationale: | |
| [insert reasoning here] | |
| ### Answer: | |
| [insert direct answer here] | |
| ``` | |
| ## Benchmarks | |
| | Model | Size | Data | Method | GSM8K (5-shot) | AGIEval (English/Nous subset, acc_norm) | BIG Bench Hard (CoT, few-shot*) | | |
| |:-----------------------------------------------------------------------|--------|:--------------------|---------------|:---------------|:----------------------------------------|:------------------------------ | | |
| | [StableLM 3B Base](https://hf.co/stabilityai/stablelm-3b-4e1t) | 3B | Base | Base | 2.05% | 25.14% | 36.75% | | |
| | [StableHermes 3B](https://hf.co/cxllin/StableHermes-3b) | 3B | GPT | SFT | 3.64% | 24.31% | **37.28%** | | |
| | [MPT 7B Instruct](https://hf.co/mosaicml/mpt-7b-instruct) | **7B** | **Human**+Anthropic | SFT | 2.05% | 24.12% | 11.01% | | |
| | [OpenLLaMA 7B v2 open-instruct](http://hf.co/VMware/open-llama-7b-v2-open-instruct) | **7B** | **Human** (nearly: ecqa is an exception) | SFT | 8.64% | 23.21% | 29.84% | | |
| | [StableLM Zephyr 3B](https://hf.co/stabilityai/stablelm-zephyr-3b) | 3B | GPT | DPO | possibly contaminated (45.72%) | **33.31%** | 0.91% | | |
| | [LIMA LLaMA 2 7B](https://huggingface.co/heegyu/LIMA2-7b-hf) | **7B** | **Human** | SFT | 4.55% | 24.55% | 36.29% | | |
| | [**Memphis-CoT 3B**](https://hf.co/euclaise/Memphis-CoT-3B) | 3B | **Human** | Self-teaching | **18.8%** | *27.22%* | *36.92%* | | |
| *5-shot, as performed automatically by LM Evaluation Harness bbh_cot_fewshot even with num_fewshot=0 | |
| Memphis outperforms other primarily-human-data models that are over twice its size, along with SFT models of its size, and trades with the Zephyr DPO model. That said, Zephyr uses synthetic data, and *much* more of it. | |
| Note that BBH results have wide SEs, sometimes even exceeding 16%. | |
| It is unclear why Zephyr performs so poorly on BBH. Perhaps it is overfit, or maybe there was an issue with vllm. | |
| Notes: | |
| - Evaluations were performed using the `agieval` branch of [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) (commit `0bef5c9c273b1c2f68e6018d4bb9c32b9aaff298`), using the `vllm` model. | |
| - I tried to find human-data-trained StableLM models, but couldn't find any. I did find a few OpenLLaMA models, but they wouldn't load with LM Eval Harness and vllm. (I believe this can be fixed by changing the xformers backend, but I'm too lazy for that) | |
| - OpenLLaMA 7B v2 open-instruct is a particularly relevant comparison, as it was trained on a *very* similar dataset. | |
| ## Hyperparameters | |
| For the initial supervised finetuning step: | |
| - Adalite optimizer, default hyperparameters of supertrainer2000 unless otherwise specified | |
| - Lambda (Adalite's analogue to weight decay, see [here](https://arxiv.org/abs/2103.06583) for details) of 0.01 | |
| - LR of 1e-5 | |
| - MixCE ratio of 0.75 | |
| - Sequence length of 4096 | |
| - Cosine decay with a 20% warmup | |
| - Frozen embeddings | |
| - No training on inputs | |
| - Accumulated batch size of 128 | |
| - NEFTune with an alpha of 10 | |
| For the generations: | |
| - Generated using the current git version of `vllm` | |
| - N=8 | |
| - Temperature of 0.5 | |
| - `top_p` of 0.8 | |
| - Maximum of 512 generated tokens, discarding responses that do not have a valid rationale and answer | |
| For the rank finetuning: | |
| - Adalite optimizer, default hyperparameters of supertrainer2000 unless otherwise specified | |
| - Lambda of 0.01 | |
| - LR of 5e-7 | |
| - Rank loss weight of 0.25 | |
| - Sequence length of 1024 | |
| - Cosine schedule with 10% warmup | |
| - Frozen embeddings | |
| - No training on inputs | |
| - Accumulated batch size of 128 | |
| - NEFTune with an alpha of 10 | |
| Additional thanks to @nicoboss for giving me access to his private supercomputer, enabling me to provide many more quants, at much higher speed, than I would otherwise be able to. |