Einstein Models
Einstein series
Text Generation • Updated • 4.04k • 34Note 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 • Updated • 378 • 66Note 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-v6.1-phi2
Text Generation • Updated • 15
Weyaxi/Einstein-v6-7B
Text Generation • Updated • 16 • 12Note 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 • Updated • 26 • 10Note ⚠️ 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 • Updated • 11 • 3Note 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 • Updated • 9 • 1Note Phi2 full fine-tuned for 2 epochs with the exact data used to train Einstein-v4-7B.
Weyaxi/Einstein-v4-7B
Text Generation • Updated • 121 • 47Note 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 • Updated • 128 • 1Note New portion of data compared to v2.
Weyaxi/Einstein-v2-7B
Text Generation • Updated • 125 • 1Note 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 • 12 • 10Note 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.