100% – I think a basic jinja implementation would be quite helpful. I do think a world where we have unified chat templates across transformers, llama.cpp, etc would be quite headache free 😀 There is an abandoned PR adding that in llama.cpp btw https://
github.com/ggerganov/llam
a.cpp/pull/9639
…
@reach_vb
-
Unified Jinja Chat Templates for LLMs and llama.cpp
By
–
-
llama.cpp Static Templates vs Jinja Support Discussion
By
–
100% yes, llama.cpp uses static templates atm and doesn't have jinja support (primarily because jinja is quite bulky). Static templates work quite well tho (Except when they don't :p) cc: @ggerganov too
-
Base Models in AI: Use Cases and Merging Strategies
By
–
Ah! Interesting, so only show `base_models`. What's the main use-case behind it (just to understand it better)? Note: it does get a bit tricky because sometimes you can have merges or adpater based models which are totally new, but just use another model as the base.
-
Cloning and Running AI Spaces Locally or on ZeroGPU
By
–
Not sure if I follow, you can just clone the space and run locally right? Alternatively you can run it on ZeroGPU
-
Model Parameters and Precision Information Now Available
By
–
Great feature request. There is a version of that in the works, just FYI you can see how many parameters does the model have on the model page and it's corresponding precision, which should give a rough idea – loads more to do there. In the meantime you can use this too:
-
GGUF Viewer Tool Enables Model Layer and Metadata Exploration
By
–
Brilliant feature request, have you seen the GGUF viewer on the Hub? It allows you to explore the layers + the metadata both. Is this close to what you envision? What else would you like to see here? (to open this, you click on the GG button next to the GGUF file – we should
-
Improving Hugging Face Hub: Community Feature Requests
By
–
If you were to fix one thing on the Hugging Face Hub, what would it be? What's your biggest gripe with the hub – let's fix it; the more, the merrier.
-
Open Source AI Narrowing Gap with Proprietary Solutions
By
–
In the end open source always catches up: we’re at < 8 month lag time now – this will only gets shorter, much shorter! 🔥 https://t.co/n1X7fHNMyx pic.twitter.com/MojF6ufukZ
— Vaibhav (VB) Srivastav (@reach_vb) 16 novembre 2024In the end open source always catches up: we’re at < 8 month lag time now – this will only gets shorter, much shorter!
-
Ultravox 0.4.1: Open-Source Audio LM Approaching GPT-4o Level
By
–
Audio LMs scene is heating up! 🔥 @FixieAI Ultravox 0.4.1 – 8B model approaching GPT4o level, pick any LLM, train an adapter with Whisper as Audio Encoder, profit 💥
— Vaibhav (VB) Srivastav (@reach_vb) 16 novembre 2024
Bonus: MIT licensed checkpoints
> Pre-trained on Llama3.1-8b/ 70b backbone as well as the encoder part of… pic.twitter.com/uB6FNJZvRbAudio LMs scene is heating up! @FixieAI Ultravox 0.4.1 – 8B model approaching GPT4o level, pick any LLM, train an adapter with Whisper as Audio Encoder, profit Bonus: MIT licensed checkpoints > Pre-trained on Llama3.1-8b/ 70b backbone as well as the encoder part of