RWKV-7 "Goose" with Expressive Dynamic State Evolution
Paper
•
2503.14456
•
Published
•
153
bndp
bndp
·
AI & ML interests
None yet
Recent Activity
upvoted
a
paper
about 1 month ago
RWKV-7 "Goose" with Expressive Dynamic State Evolution
liked
a Space
about 1 month ago
abidlabs/EasyGhibli
liked
a model
about 1 month ago
POLARIS-Project/Polaris-4B-Preview
Organizations
None yet
upvoted
a
paper
about 1 month ago
liked
a
Space
about 1 month ago
liked
a
model
about 1 month ago
reacted to
alandao's
post with 🔥
about 1 month ago
Post
1238
Don’t give up 🔥
Do you know what I was planning to do this time last week?
I was preparing to write a report declaring that Jan Nano was a failed project because the benchmark results didn’t meet expectations.
But I thought — it can’t be. When loading the model into the app, the performance clearly felt better. So why were the benchmark results worse?
That’s when I reviewed the entire benchmark codebase and realized something fundamental: agentic or workflow-based approaches introduce a huge gap and variation when benchmarking. Jan-nano was trained with an agentic setup — it simply can’t be benchmarked using a rigid workflow-based method.
I made the necessary changes, and the model ended up performing even better than before the issues arose. Turns out the previous benchmarking method conflicted with the way the model was trained.
What if I had given up? That would’ve meant 1.5 months of training and a huge amount of company resources wasted.
But now, this is officially the most successful and biggest release for the whole team — all thanks to Jan-nano.
Menlo/Jan-nano
Do you know what I was planning to do this time last week?
I was preparing to write a report declaring that Jan Nano was a failed project because the benchmark results didn’t meet expectations.
But I thought — it can’t be. When loading the model into the app, the performance clearly felt better. So why were the benchmark results worse?
That’s when I reviewed the entire benchmark codebase and realized something fundamental: agentic or workflow-based approaches introduce a huge gap and variation when benchmarking. Jan-nano was trained with an agentic setup — it simply can’t be benchmarked using a rigid workflow-based method.
I made the necessary changes, and the model ended up performing even better than before the issues arose. Turns out the previous benchmarking method conflicted with the way the model was trained.
What if I had given up? That would’ve meant 1.5 months of training and a huge amount of company resources wasted.
But now, this is officially the most successful and biggest release for the whole team — all thanks to Jan-nano.
Menlo/Jan-nano
published
a
model
about 2 months ago
updated
a
model
about 2 months ago
reacted to
AdinaY's
post with 🔥
about 2 months ago
Post
2687
RedNote 小红书 just released their first LLM 🔥
dots.llm1.base 🪐 a 142B MoE model with only 14B active params.
rednote-hilab/dotsllm1-68246aaaaba3363374a8aa7c
✨ Base & Instruct - MIT license
✨ Trained on 11.2T non-synthetic high-quality data
✨ Competitive with Qwen2.5/3 on reasoning, code, alignment
dots.llm1.base 🪐 a 142B MoE model with only 14B active params.
rednote-hilab/dotsllm1-68246aaaaba3363374a8aa7c
✨ Base & Instruct - MIT license
✨ Trained on 11.2T non-synthetic high-quality data
✨ Competitive with Qwen2.5/3 on reasoning, code, alignment
liked
2
Spaces
about 2 months ago
liked
a
Space
2 months ago
reacted to
csabakecskemeti's
post with 👍
4 months ago
Post
3380
I'm collecting llama-bench results for inference with a llama 3.1 8B q4 and q8 reference models on varoius GPUs. The results are average of 5 executions.
The system varies (different motherboard and CPU ... but that probably that has little effect on the inference performance).
https://devquasar.com/gpu-gguf-inference-comparison/
the exact models user are in the page
I'd welcome results from other GPUs is you have access do anything else you've need in the post. Hopefully this is useful information everyone.
The system varies (different motherboard and CPU ... but that probably that has little effect on the inference performance).
https://devquasar.com/gpu-gguf-inference-comparison/
the exact models user are in the page
I'd welcome results from other GPUs is you have access do anything else you've need in the post. Hopefully this is useful information everyone.
upvoted
an
article
4 months ago
Article
Open R1: How to use OlympicCoder locally for coding?
By
and 4 others
•
•
62
reacted to
AdinaY's
post with 🔥
5 months ago
Post
2927
RWKV7-G1 0.1B 🔥 Pure RNN reasoning model released by RWKV
Model: BlinkDL/rwkv7-g1
paper: RWKV-7 "Goose" with Expressive Dynamic State Evolution (2503.14456)
✨ Apache2.0
✨ Supports 100+ languages
✨ 0.1 B runs smoothly on low power devices
✨ 0.4B/1.5B/2.9B are coming soon!!
Model: BlinkDL/rwkv7-g1
paper: RWKV-7 "Goose" with Expressive Dynamic State Evolution (2503.14456)
✨ Apache2.0
✨ Supports 100+ languages
✨ 0.1 B runs smoothly on low power devices
✨ 0.4B/1.5B/2.9B are coming soon!!
- 1 reply
liked
a
model
5 months ago
reacted to
AtAndDev's
post with 👍
5 months ago
Post
4347
There seems to multiple paid apps shared here that are based on models on hf, but some ppl sell their wrappers as "products" and promote them here. For a long time, hf was the best and only platform to do oss model stuff but with the recent AI website builders anyone can create a product (really crappy ones btw) and try to sell it with no contribution to oss stuff. Please dont do this, or try finetuning the models you use...
Sorry for filling yall feed with this bs but yk...
Sorry for filling yall feed with this bs but yk...
- 6 replies
reacted to
BlinkDL's
post with 🔥
5 months ago
Post
11208
RWKV-6-world-v3 (+3.1T tokens) is our best multilingual 7B model as of now:
BlinkDL/rwkv-6-world
It's 100% RNN and attention-free. MMLU 54.2% (previous world-v2.1 = 47.9%. note: without eval-boosting tricks such as annealing).
RWKV-7-world-v4 soon :)
It's 100% RNN and attention-free. MMLU 54.2% (previous world-v2.1 = 47.9%. note: without eval-boosting tricks such as annealing).
RWKV-7-world-v4 soon :)
- 1 reply
upvoted
a
collection
5 months ago
liked
a
Space
5 months ago
upvoted
a
collection
5 months ago
reacted to
fdaudens's
post with 👍
5 months ago
Post
4125
AI will bring us "a country of yes-men on servers" instead of one of "Einsteins sitting in a data center" if we continue on current trends.
Must-read by @thomwolf deflating overblown AI promises and explaining what real scientific breakthroughs require.
https://thomwolf.io/blog/scientific-ai.html
Must-read by @thomwolf deflating overblown AI promises and explaining what real scientific breakthroughs require.
https://thomwolf.io/blog/scientific-ai.html
- 2 replies