Closed neokd closed 7 months ago
Assign me under JWoC, but before that could you provide me with the reference for the cost for models.
As here https://www.together.ai/pricing and https://docs.together.ai/docs/inference-models , its hard to understand the notation on how i should write the json key
TOGETHERAI_MODEL_COST_PER_1M_TOKENS = {
"mistralai/Mistral-7B-Instruct-v0.2": 0.20,
"mistralai/Mixtral-8x7B-Instruct-v0.1": 0.60,
}
If you sign up for together AI. On the models page you will find the model name and it's pricing. You can get get it from the table.
Here's the updated JSON with Chat, Language and Code models of Together.ai Review it beforehand so i can commit.
TOGETHERAI_MODEL_COST_PER_1M_TOKENS = {
"deepseek-ai/deepseek-coder-33b-instruct":0.80,
"Qwen/Qwen1.5-72B-Chat":0.90,
"Qwen/Qwen1.5-14B-Chat":0.30,
"Qwen/Qwen1.5-7B-Chat":0.20,
"Qwen/Qwen1.5-4B-Chat":0.10,
"Qwen/Qwen1.5-1.8B-Chat":0.10,
"Qwen/Qwen1.5-0.5B-Chat":0.10,
"codellama/CodeLlama-70b-Instruct-hf": 0.90,
"mistralai/Mixtral-8x7B-Instruct-v0.1": 0.60,
"mistralai/Mistral-7B-Instruct-v0.2": 0.20,
"meta-llama/Llama-2-70b-chat-hf": 0.90,
"snorkelai/Snorkel-Mistral-PairRM-DPO": 0.20,
"codellama/CodeLlama-13b-Instruct-hf": 0.22,
"codellama/CodeLlama-34b-Instruct-hf": 0.776,
"codellama/CodeLlama-7b-Instruct-hf": 0.20,
"NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO": 0.60,
"NousResearch/Nous-Hermes-2-Mixtral-8x7B-SFT": 0.60,
"NousResearch/Nous-Hermes-2-Yi-34B": 0.80,
"openchat/openchat-3.5-1210": 0.20,
"togethercomputer/StripedHyena-Nous-7B": 0.20,
"DiscoResearch/DiscoLM-mixtral-8x7b-v2": 0.60,
"mistralai/Mistral-7B-Instruct-v0.1": 0.20,
"zero-one-ai/Yi-34B-Chat": 0.80,
"NousResearch/Nous-Capybara-7B-V1p9": 0.20,
"teknium/OpenHermes-2p5-Mistral-7B": 0.20,
"upstage/SOLAR-10.7B-Instruct-v1.0": 0.30,
"togethercomputer/llama-2-13b-chat": 0.22,
"togethercomputer/llama-2-7b-chat": 0.20,
"NousResearch/Nous-Hermes-Llama2-13b": 0.30,
"NousResearch/Nous-Hermes-llama-2-7b": 0.20,
"Open-Orca/Mistral-7B-OpenOrca": 0.20,
"teknium/OpenHermes-2-Mistral-7B": 0.20,
"WizardLM/WizardLM-13B-V1.2": 0.20,
"togethercomputer/Llama-2-7B-32K-Instruct": 0.20,
"lmsys/vicuna-13b-v1.5": 0.30,
"Austism/chronos-hermes-13b": 0.30,
"garage-bAInd/Platypus2-70B-instruct": 0.90,
"Gryphe/MythoMax-L2-13b": 0.30,
"togethercomputer/Qwen-7B-Chat": 0.20,
"togethercomputer/RedPajama-INCITE-7B-Chat": 0.20,
"togethercomputer/RedPajama-INCITE-Chat-3B-v1": 0.10,
"togethercomputer/alpaca-7b": 0.20,
"togethercomputer/falcon-40b-instruct": 0.80,
"togethercomputer/falcon-7b-instruct": 0.20,
"Qwen/Qwen1.5-72B": 0.90,
"Qwen/Qwen1.5-14B": 0.30,
"Qwen/Qwen1.5-7B": 0.20,
"Qwen/Qwen1.5-4B": 0.10,
"Qwen/Qwen1.5-1.8B": 0.10,
"Qwen/Qwen1.5-0.5B": 0.10,
"mistralai/Mixtral-8x7B-v0.1": 0.60,
"meta-llama/Llama-2-70b-hf": 0.90,
"togethercomputer/StripedHyena-Hessian-7B": 0.20,
"mistralai/Mistral-7B-v0.1": 0.20,
"microsoft/phi-2": 0.10,
"zero-one-ai/Yi-34B": 0.80,
"zero-one-ai/Yi-6B": 0.14,
"Nexusflow/NexusRaven-V2-13B": 0.30,
"togethercomputer/LLaMA-2-7B-32K": 0.20,
"togethercomputer/llama-2-13b": 0.22,
"togethercomputer/llama-2-7b": 0.20,
"togethercomputer/Qwen-7B": 0.20,
"togethercomputer/RedPajama-INCITE-7B-Instruct": 0.20,
"togethercomputer/RedPajama-INCITE-7B-Base": 0.20,
"togethercomputer/RedPajama-INCITE-Instruct-3B-v1": 0.10,
"togethercomputer/RedPajama-INCITE-Base-3B-v1": 0.10,
"togethercomputer/GPT-JT-Moderation-6B": 0.20,
"togethercomputer/falcon-40b": 0.80,
"togethercomputer/falcon-7b": 0.20,
"codellama/CodeLlama-70b-Python-hf": 0.90,
"codellama/CodeLlama-70b-hf": 0.90,
"codellama/CodeLlama-13b-Python-hf": 0.22,
"codellama/CodeLlama-34b-Python-hf": 0.776,
"codellama/CodeLlama-7b-Python-hf": 0.20,
"WizardLM/WizardCoder-Python-34B-V1.0": 0.80,
"Phind/Phind-CodeLlama-34B-v2": 0.80,
}
How many models are there?
Around 77 (including the previous to mistral ai models which were already there in the JSON) and excluding the Image Models, cuz the Cost was variable.
Also, is there any default value for --MAX-BUDGET
flag in main.py
no by default let it be none
@ayushmorbar any updates on this?
Yeah wait, I am about to commit the code in a while, was studying for the exams~
Thik hai
Done.
Update the JSON with cost for models in TogetherAI in callback_handler.py and add a flag called
--max-budget
in the main.py.