Model merging lessons in The Waifu Research Department
January 29, 2024
19 min
6 of 156

January 29, 2024
Model merging lessons in The Waifu Research Department
0:00 / 19:04
Excerpt playback isn’t available for this episode — use the full-episode link.
Search the Record
Moment search isn’t open for this episode yet
When it opens, you’ll be able to type any phrase and jump straight to the second it was said.
Nathan Lambert Episodes Around January 29, 2024
Episode 6 of 156 — walk the feed in the order it was published.
19:04This episodeModel merging lessons in The Waifu Research Department
Summary
<p>Note: some of the audio in the second half is a little wonky, but the general voice was upgraded so hopefully it's a little less "poppy" until then!<br />I'm trying to fix little pronunciation problems on a weekly basis. Thanks to my early fans! It'll keep improving. E.g. some of the months were wonky.</p><p>When what seems like pure LLM black magic is actually supported by the literature.<br />This is AI generated audio with Python and 11Labs<br />Source code: <a class="linkified" href="https://github.com/natolambert/interconnects-tools" target="_blank">https://github.com/natolambert/interconnects-tools</a><br />Original post: <a class="linkified" href="https://www.interconnects.ai/p/model-merging" target="_blank">https://www.interconnects.ai/p/model-merging</a></p><p>00:00 Model merging lessons in The Waifu Research Department<br />02:21 How and why does model merging work?<br />07:13 Aside: merging vs. ensembles vs. mixture of experts<br />08:21 Why are people doing this?<br />11:22 Tools & Links<br />11:51 Brief (visual) literature review<br />12:07 Full model merging and recent methods<br />15:55 Weight averaging during pretraining<br />17:18 LoRA merging<br />17:53 More background</p><p>Figure 1: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_005.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_005.png</a><br />Figure 2: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_016.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_016.png</a><br />Figure 3: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_042.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_042.png</a><br />Figure 4: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_051.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_051.png</a><br />Figure 5: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_055.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_055.png</a><br />Figure 6: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_058.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_058.png</a><br />Figure 7: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_060.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_060.png</a><br />Figure 8: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_062.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_062.png</a><br />Figure 9: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_065.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_065.png</a><br />Figure 10: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_075.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_075.png</a><br />Figure 11: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_077.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_077.png</a><br />Figure 12: <a class="linkified" href="https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_084.png" target="_blank">https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_084.png</a></p> <br /><br />This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit <a href="https://www.interconnects.ai/subscribe?utm_medium=podcast&utm_campaign=CTA_2">www.interconnects.ai/subscribe</a>



















