Skip to content

Update SQ/WOQ status #1869

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jun 14, 2024
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
266 changes: 132 additions & 134 deletions docs/source/llm_recipes.md
Original file line number Diff line number Diff line change
Expand Up @@ -29,8 +29,8 @@ This document aims to publish the specific recipes we achieved for the popular L
| databricks/dolly-v2-12b | ✖ | ✔ | ✖ |
| EleutherAI/gpt-neox-20b | ✖ | ✔ | ✔ |
| mistralai/Mistral-7B-v0.1 | ✖ | ✔ | ✔ |
| THUDM/chatglm2-6b | WIP | ✔ | ✔ |
| THUDM/chatglm3-6b | WIP | ✔ | |
| THUDM/chatglm2-6b | | ✔ | ✔ |
| THUDM/chatglm3-6b | WIP | ✔ | WIP |

**Detail recipes can be found [HERE](https://github.com/intel/intel-extension-for-transformers/blob/main/examples/huggingface/pytorch/text-generation/quantization/llm_quantization_recipes.md).**

Expand All @@ -40,8 +40,8 @@ This document aims to publish the specific recipes we achieved for the popular L
> - The WIP recipes will be published soon.

## Large Language Models Accuracy
<table>
<thead>

<table><thead>
<tr>
<th rowspan="3">Model</th>
<th colspan="9">lambada_openai</th>
Expand All @@ -63,212 +63,210 @@ This document aims to publish the specific recipes we achieved for the popular L
<th>Ratio</th>
<th>ACC</th>
<th>Ratio</th>
</tr>
</thead>
</tr></thead>
<tbody>
<tr>
<td>baichuan-inc/Baichuan-13B-Chat</td>
<td>67.57%</td>
<td>68.23%</td>
<td>1.0098</td>
<td>67.57%</td>
<td>1.0000</td>
<td>67.84%</td>
<td>1.0040</td>
<td>NA</td>
<td>NA</td>
<td>69.07%</td>
<td>1.0222</td>
<td>67.55%</td>
<td>0.9997</td>
<td>68.12%</td>
<td>1.0081</td>
<td>66.93%</td>
<td>0.9905</td>
</tr>
<tr>
<td>baichuan-inc/Baichuan2-13B-Chat</td>
<td>71.51%</td>
<td>70.89%</td>
<td>0.9913</td>
<td>71.53%</td>
<td>1.0003</td>
<td>71.76%</td>
<td>1.0035</td>
<td>NA</td>
<td>NA</td>
<td>75.57%</td>
<td>1.0568</td>
<td>71.57%</td>
<td>1.0008</td>
<td>70.81%</td>
<td>0.9902</td>
<td>N/A</td>
<td>N/A</td>
</tr>
<tr>
<td>baichuan-inc/Baichuan2-7B-Chat</td>
<td>67.67%</td>
<td>67.96%</td>
<td>1.0043</td>
<td>67.59%</td>
<td>0.9988</td>
<td>67.24%</td>
<td>0.9936</td>
<td>67.42%</td>
<td>0.9963</td>
<td>68.06%</td>
<td>1.0058</td>
<td>67.61%</td>
<td>0.9991</td>
<td>67.90%</td>
<td>1.0034</td>
<td>N/A</td>
<td>N/A</td>
</tr>
<tr>
<td>bigscience/bloom-1b7</td>
<td>46.34%</td>
<td>47.99%</td>
<td>1.0356</td>
<td>46.38%</td>
<td>1.0009</td>
<td>46.19%</td>
<td>0.9968</td>
<td>NA</td>
<td>NA</td>
<td>46.21%</td>
<td>0.9972</td>
<td>46.90%</td>
<td>1.0121</td>
<td>N/A</td>
<td>N/A</td>
</tr>
<tr>
<td>databricks/dolly-v2-12b</td>
<td>64.35%</td>
<td>NA</td>
<td>NA</td>
<td>64.10%</td>
<td>0.9961</td>
<td>NA</td>
<td>NA</td>
<td>NA</td>
<td>NA</td>
<td>N/A</td>
<td>N/A</td>
<td>63.92%</td>
<td>0.9933</td>
<td>N/A</td>
<td>N/A</td>
<td>N/A</td>
<td>N/A</td>
</tr>
<tr>
<td>EleutherAI/gpt-j-6b</td>
<td>68.31%</td>
<td>68.33%</td>
<td>1.0003</td>
<td>68.23%</td>
<td>0.9988</td>
<td>68.79%</td>
<td>1.0070</td>
<td>68.43%</td>
<td>1.0018</td>
<td>68.27%</td>
<td>0.9994</td>
<td>68.27%</td>
<td>0.9994</td>
<td>68.35%</td>
<td>1.0006</td>
<td>68.02%</td>
<td>0.9958</td>
</tr>
<tr>
<td>EleutherAI/gpt-neox-20b</td>
<td>72.33%</td>
<td>NA</td>
<td>NA</td>
<td>72.25%</td>
<td>0.9989</td>
<td>71.96%</td>
<td>0.9949</td>
<td>NA</td>
<td>NA</td>
<td>N/A</td>
<td>N/A</td>
<td>72.29%</td>
<td>0.9994</td>
<td>71.74%</td>
<td>0.9918</td>
<td>N/A</td>
<td>N/A</td>
</tr>
<tr>
<td>facebook/opt-1.3b</td>
<td>57.89%</td>
<td>57.54%</td>
<td>0.9940</td>
<td>58.08%</td>
<td>1.0033</td>
<td>58.57%</td>
<td>1.0117</td>
<td>NA</td>
<td>NA</td>
<td>57.68%</td>
<td>0.9964</td>
<td>58.12%</td>
<td>1.0040</td>
<td>58.26%</td>
<td>1.0064</td>
<td>N/A</td>
<td>N/A</td>
</tr>
<tr>
<td>facebook/opt-30b</td>
<td>71.49%</td>
<td>71.51%</td>
<td>1.0003</td>
<td>71.51%</td>
<td>1.0003</td>
<td>71.82%</td>
<td>1.0046</td>
<td>72.11%</td>
<td>1.0087</td>
<td>71.78%</td>
<td>1.0041</td>
<td>71.53%</td>
<td>1.0006</td>
<td>71.59%</td>
<td>1.0014</td>
<td>71.80%</td>
<td>1.0043</td>
</tr>
<tr>
<td>meta-llama/Llama-2-13b-hf</td>
<td>76.77%</td>
<td>76.25%</td>
<td>0.9932</td>
<td>76.75%</td>
<td>0.9997</td>
<td>77.43%</td>
<td>1.0086</td>
<td>76.75%</td>
<td>0.9997</td>
<td>76.89%</td>
<td>1.0016</td>
<td>77.66%</td>
<td>1.0116</td>
<td>76.60%</td>
<td>0.9978</td>
</tr>
<tr>
<td>meta-llama/Llama-2-70b-hf</td>
<td>79.64%</td>
<td>79.55%</td>
<td>0.9989</td>
<td>79.57%</td>
<td>0.9991</td>
<td>79.14%</td>
<td>0.9937</td>
<td>79.62%</td>
<td>0.9997</td>
<td>80.09%</td>
<td>1.0057</td>
<td>79.97%</td>
<td>1.0041</td>
<td>79.68%</td>
<td>1.0005</td>
</tr>
<tr>
<td>meta-llama/Llama-2-7b-hf</td>
<td>73.92%</td>
<td>73.45%</td>
<td>0.9936</td>
<td>73.96%</td>
<td>1.0005</td>
<td>73.45%</td>
<td>0.9936</td>
<td>73.49%</td>
<td>0.9942</td>
<td>73.90%</td>
<td>0.9997</td>
<td>73.84%</td>
<td>0.9989</td>
<td>N/A</td>
<td>N/A</td>
</tr>
<tr>
<td>mistralai/Mistral-7B-v0.1</td>
<td>75.90%</td>
<td>NA</td>
<td>NA</td>
<td>N/A</td>
<td>N/A</td>
<td>75.80%</td>
<td>0.9987</td>
<td>76.13%</td>
<td>1.0030</td>
<td>75.61%</td>
<td>0.9962</td>
<td>76.25%</td>
<td>1.0046</td>
<td>75.74%</td>
<td>0.9979</td>
</tr>
<tr>
<td>THUDM/chatglm2-6b</td>
<td>53.23%</td>
<td>NA</td>
<td>NA</td>
<td>53.19%</td>
<td>0.9992</td>
<td>52.77%</td>
<td>0.9914</td>
<td>53.35%</td>
<td>1.0023</td>
<td>52.86%</td>
<td>0.9930</td>
<td>53.00%</td>
<td>0.9957</td>
<td>52.90%</td>
<td>0.9938</td>
<td>52.92%</td>
<td>0.9942</td>
</tr>
<tr>
<td>THUDM/chatglm3-6b</td>
<td>59.09%</td>
<td>NA</td>
<td>NA</td>
<td>59.01%</td>
<td>0.9986</td>
<td>NA</td>
<td>NA</td>
<td>58.61%</td>
<td>0.9919</td>
<td>N/A</td>
<td>N/A</td>
<td>59.03%</td>
<td>0.9990</td>
<td>N/A</td>
<td>N/A</td>
<td>N/A</td>
<td>N/A</td>
</tr>
<tr>
<td>tiiuae/falcon-40b</td>
<td>77.22%</td>
<td>77.04%</td>
<td>0.9977</td>
<td>77.22%</td>
<td>1.0000</td>
<td>77.94%</td>
<td>1.0093</td>
<td>78.79%</td>
<td>1.0203</td>
<td>76.95%</td>
<td>0.9965</td>
<td>77.18%</td>
<td>0.9995</td>
<td>77.55%</td>
<td>1.0043</td>
<td>77.82%</td>
<td>1.0078</td>
</tr>
<tr>
<td>tiiuae/falcon-7b</td>
<td>74.67%</td>
<td>76.44%</td>
<td>1.0237</td>
<td>74.77%</td>
<td>1.0013</td>
<td>75.00%</td>
<td>1.0044</td>
<td>NA</td>
<td>NA</td>
<td>76.63%</td>
<td>1.0262</td>
<td>74.73%</td>
<td>1.0008</td>
<td>75.06%</td>
<td>1.0052</td>
<td>74.00%</td>
<td>0.9910</td>
</tr>
</tbody>
</table>
</tbody></table>
Loading