google/gemma-2-27b-it

Text Generation • Updated Aug 27 • 199k • 471

Note This list is not very stable thanks to Google, new local function calling king is here! Gets all the required functions (6/6) for the example eval: https://github.com/AtakanTekparmak/Prompts/blob/main/function_calling/chained_fnc_hard.txt

google/gemma-2-9b-it

Text Generation • Updated Aug 27 • 344k • 592

Note Google isn't playing with the Gemma-2 release, gets 5/6 out of the required functions (1 is not a reasoning error but type error). Runs fast, is good, what more do you want

01-ai/Yi-1.5-34B-Chat

Text Generation • Updated Aug 27 • 10k • • 258

Note Great for multi-function calling, reasoning, Chain of Thought and planning, a bit slow given the memory limit but best general-purpose full-precision model one can use in these constraints, great Apache 2.0 license also. Gets 4/6 required functions in the example prompt.

THUDM/glm-4-9b-chat

Updated 26 days ago • 490k • 646

Note P4P (Parameter-for-Parameter) best local function calling open-weights model, which can be run full-precision in less than 64GB of memory, as of today (June 6 2024). On par with GPT-4-Turbo on Berkeley Function Calling Leaderboard (https://huggingface.co/THUDM/glm-4-9b-chat#%E5%B7%A5%E5%85%B7%E8%B0%83%E7%94%A8%E8%83%BD%E5%8A%9B). Gets 4/6 in the example prompt.

NousResearch/Hermes-2-Pro-Llama-3-8B

Text Generation • Updated Sep 14 • 23.8k • 413

Note Hermes fine-tune for the best <10B Llama released to date, does well on everything, specifically function calling which it was specifically fine-tuned on. Gets 4/6 in the example prompt.

google/gemma-1.1-7b-it

Text Generation • Updated Jun 27 • 15.7k • • 267

Note Surprisingly decent for the size, very good at instruction following / adhering to the format. Also does relatively well on multiple function calling, struggles after 3 chained functions, getting only 3/6 in the example prompt.