Weyaxi/Einstein-v7-Qwen2-7B

Text Generation • 8B • Updated Jul 23, 2024 • 27 • 42

Note Based on Qwen2 7B. Full fine-tuned for 2 epochs. Version 7 included more data. Added systemchat 1.1 and some of the Buzz dataset.

Weyaxi/Einstein-v6.1-Llama3-8B

Text Generation • 8B • Updated Jul 23, 2024 • 8.07k • 70

Note Currently the best and stable model. Based on the new LLama3. Full fine-tuned for 2 epochs. Version 6 included more data and changed some of the datasets that has been used in v5. Added EverythingLM-data-V3, no_robots and half of the wizardlm_evol_instruct_70k dataset. Version 6.1 removed refusals etc. to remove censorship compared to v6.

Weyaxi/Einstein-v6-7B

Text Generation • 7B • Updated Apr 11, 2024 • 14 • 12

Note Good and stable model but it's more censored than v4. Based on the fixed version of Mistral-7B-v0.2 Full fine-tuned for 2 epochs. Included more data and changed some of the datasets I used. Added EverythingLM-data-V3, no_robots and half of the wizardlm_evol_instruct_70k dataset.

Weyaxi/Einstein-v5-v0.2-7B

Text Generation • 7B • Updated Apr 4, 2024 • 14 • 10

Note ⚠️ Deprecated! Based on Mistral-7B-v0.2 but fine-tuned with wrong sliding window. Full fine-tuned for 1 epochs. Included more data and changed some of the datasets I used.

Weyaxi/Einstein-v4-Qwen-1.5-32B

Text Generation • 33B • Updated Jun 15, 2024 • 17 • 3

Note Based on Qwen1.5-32B, qlora fine-tuned using 2 epochs with the exact data used to train Einstein-v4-7B

Weyaxi/Einstein-v4-phi2

Text Generation • 3B • Updated Apr 23, 2024 • 23 • 1

Note Phi2 full fine-tuned for 2 epochs with the exact data used to train Einstein-v4-7B.

Weyaxi/Einstein-v4-7B

Text Generation • 7B • Updated Jul 23, 2024 • 77 • 48

Note Good and a stable model. It has been full fine-tuned for 1.5 epochs. Included more data and changed some of the datasets I used.

Weyaxi/Einstein-v3-7B

Text Generation • 7B • Updated Apr 4, 2024 • 19 • 1

Note New portion of data compared to v2.

Weyaxi/Einstein-v2-7B

Text Generation • 7B • Updated Apr 4, 2024 • 13 • 1

Note Test the model after the initial model. I have now begun to include both general data and science-related datasets in this model.

Weyaxi/Einstein-7B

Text Generation • Updated Jan 28, 2024 • 20 • 10

Note First model of the series. The initial idea was to train it exclusively with science-related datasets. However, I later modified this idea to prioritize having more science-related data than other existing good models.