4-bit Adapter Alignment-handbook Base model:adapter:microsoft/p... Base model:microsoft/phi-2 Bitsandbytes Custom code Dataset:huggingfaceh4/ultracha... Finetuned Generated from trainer Instruct Lora Peft Phi Region:us Safetensors
Phi 2 Instruction Benchmarks
ARC: 61.09-36.6%
HellaSwag: 74.68-21.6%
MMLU: 57.81-33.2%
TruthfulQA: 45.1-23.6%
WinoGrande: 74.82-14.5%
GSM8K: 51.63-43.9%
nn.n% — How the model compares to the GPT-4.
Phi 2 Instruction Parameters and Internals
| LLM Name | Phi 2 Instruction |
| Repository | Open on 🤗 |
| Base Model(s) | |
| Required VRAM | 0 GB |
| Updated | 2024-07-26 |
| Maintainer | huseyinatahaninan |
| Instruction-Based | Yes |
| Model Files | |
| Model Architecture | Adapter |
| License | mit |
| Model Max Length | 2048 |
| Is Biased | none |
| Tokenizer Class | CodeGenTokenizer |
| Padding Token | <|endoftext|> |
| PEFT Type | LORA |
| LoRA Model | Yes |
| PEFT Target Modules | q_proj|gate_proj|down_proj|v_proj|up_proj|k_proj|o_proj |
| LoRA Alpha | 16 |
| LoRA Dropout | 0.05 |
| R Param | 16 |
Best Alternatives to Phi 2 Instruction
Best Alternatives |
HF Rank |
Context/RAM |
Downloads |
Likes |
|---|---|---|---|---|
| Vfgf | 0.2 | 0K / 0 GB | 19 | 0 |
| Test | 0.2 | 0K / 0 GB | 13 | 0 |
| Results | 0.2 | 0K / 0 GB | 7 | 0 |
| Phi 3 Mini QLoRA | 0.2 | 0K / 0 GB | 151 | 0 |
| Phi 3 Mini 4K Instruct Ru Lora | 0.2 | 0K / 0.1 GB | 19 | 0 |
| Results Phi3 Medium 4k | 0.2 | 0K / 0.1 GB | 5 | 0 |
| Results | 0.2 | 0K / 0.1 GB | 7 | 0 |
| Phi 3 Mini 4K Instruct Sft | 0.2 | 0K / 0 GB | 12 | 0 |
| Phi3AdapterModel | 0.2 | 0K / 0.1 GB | 7 | 0 |
| Phi3 Mini 4K Qlora Adapter | 0.2 | 0K / 0 GB | 31 | 0 |
Note: green Score (e.g. "73.2") means that the model is better than huseyinatahaninan/phi-2-instruction .
Rank the Phi 2 Instruction Capabilities
🆘 Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! 🌟
What open-source LLMs or SLMs are you in search of? 34446 in total.
Original data from HuggingFace, OpenCompass and various public git repos.
Release v2024072501