Nov 4 10 days ago 03:43 Upload nemo-44khz-nar-flow-225m/fp16.sft May 22, 2025 18:26 Upload fp32.sft Reverted to prior revision Apr 19, 2025 19:03 Upload fp32.sft I hope this isn't the copy that managed to botch itself because I didn't export it the right way 01:42 Upload fp32.sft Apr 6, 2025 03:36 Upload fp32.sft Still not good enough... Apr 1, 2025 01:29 Upload fp32.sft Compared to the nemo models this is well more than adequate...... Mar 30, 2025 14:52 Upload fp32.sft 14:43 Upload fp32.sft 05:00 Upload fp32.sft Mar 29, 2025 00:18 m mrq
00:08 Upload 3 files Mar 20, 2025 01:46 Upload 2 files Mar 15, 2025 01:31 Upload 2 files Mar 11, 2025 20:49 Upload 2 files Experimental weights Jan 6, 2025 06:09 Upload fp32.sft experimental weights Dec 28, 2024 00:14 Upload fp32.sft More post-training on NS/SR that wasn't necessary but might as well push these weights......... Dec 20, 2024 15:03 Upload fp32.sft Dec 17, 2024 01:22 Upload fp32.sft Dec 16, 2024 14:36 Upload fp32.sft Dec 15, 2024 23:18 cleanup mrq
22:59 Upload fp32.sft Separating from the confusing ar+nar-llama-8 name since this is now the nar-len too. Nov 23, 2024 19:56 Update models/config.llama.yaml 19:56 Update models/config.llama.yaml 01:44 Upload fp32.sft Decent weights Nov 20, 2024 00:52 Upload 2 files Updated models (ar+nar has nar-len compatibility, nar-len model is finalized as the reference model will receive training) Nov 18, 2024 03:42 Upload fp32.sft ugh Nov 17, 2024 16:13 Upload 2 files Updated reference AR+NAR model to a "better" trained one (need to do actual regression tests, but it sounds really nice), backed up current one as `ar+nar-old-llama-8` 14:53 Upload fp32.sft Slightly better but requires more training. Nov 16, 2024 21:49 Upload 2 files Very, very early NAR-len model (that works enough but not good enough so far) Nov 5, 2024 04:48 Upload fp32.sft Again, LayerSkip is very sensitive to de-facto batch size. Nov 3, 2024 17:13 Upload fp32.sft More promising layerskip-aware weights (it actually seems to perform better than the base model, atlhough with just one voice I'm comparing so far) 02:37 Upload fp32.sft Updated layerskip weights (not looking good...) Nov 2, 2024 16:40 Upload fp32.sft Better-ish layerskip-aware weights (apparently the model didn't train as much as I would have liked overnight but it seems fine-er-ish) 02:30 Upload 2 files Very early layerskip-aware weights Oct 26, 2024 01:27 Update models/config.llama.yaml Reduced a lot of unneeded settings in the config YAML Oct 17, 2024 22:10 Update models/config.retnet.yaml Modified keys 22:09 Update models/config.llama.yaml Modified keys Oct 6, 2024 01:39 replaced ar+nar-tts+stt-llama-8 as the base ar+nar-llama-8 since it pretty much surpasses it now mrq
Oct 5, 2024 21:48 Upload fp32.sft Updated weights (post-training to try and clean up the NAR, probably sounds better?) Sep 28, 2024 13:46 Upload fp32.sft Updated weights, again (more post-training, seemed to really need it). Sep 27, 2024 04:00 Upload fp32.sft Yet another updated weights (did some more post training on sampling per group, then culled the dataset to focus more on my own corpus to help bolster it), seems to help? Sep 26, 2024 23:24 Upload fp32.sft Updated weights (received more duration-post-fix training, seems better now) 03:01 Upload fp32.sft Updated weights (the prior ones were "post-fix trained" with an out of date tokenizer) Sep 25, 2024 00:52 Upload 2 files Updated "foundational" model. Sep 8, 2024 23:00 Upload 8 files 02:02 Upload 2 files Sep 2, 2024 16:02 Upload fp32.sft Received additional training on my 7900XTX (AMP bfloat16, SDPA):
* a number of unshuffled samples from 12 seconds to 24 seconds, to verify that my ROCm setup works.
* another number of shuffled samples from 3 seconds to 32 seconds, to re-"teach" the model to work for any duration rather than the last duration it was trained against.
* another number of shuffled samples from 3 seconds to 60 seconds but with a RVQ distribution favoring the higher levels, to see if it lobotomizes the AR and clean up the NAR's audio (it seems fine). Aug 5, 2024 04:19 ugh mrq
Aug 4, 2024 05:01 Upload 2 files Updating latest copies in safetensor form. Jul 31, 2024 03:45 Upload 3 files 
