diff options
author | kqlio67 <166700875+kqlio67@users.noreply.github.com> | 2024-07-29 00:37:16 +0200 |
---|---|---|
committer | kqlio67 <166700875+kqlio67@users.noreply.github.com> | 2024-07-29 00:37:16 +0200 |
commit | 51faa61f236b90efcb865eda1eb07b04e7416fc6 (patch) | |
tree | 1529bbfeb3ffc3ba9f3124e6987a2e6fc6ef2ac6 /g4f | |
parent | Added Marsyoo provider with support for the gpt-4o model. Updating models.py (diff) | |
download | gpt4free-51faa61f236b90efcb865eda1eb07b04e7416fc6.tar gpt4free-51faa61f236b90efcb865eda1eb07b04e7416fc6.tar.gz gpt4free-51faa61f236b90efcb865eda1eb07b04e7416fc6.tar.bz2 gpt4free-51faa61f236b90efcb865eda1eb07b04e7416fc6.tar.lz gpt4free-51faa61f236b90efcb865eda1eb07b04e7416fc6.tar.xz gpt4free-51faa61f236b90efcb865eda1eb07b04e7416fc6.tar.zst gpt4free-51faa61f236b90efcb865eda1eb07b04e7416fc6.zip |
Diffstat (limited to 'g4f')
-rw-r--r-- | g4f/gui/client/index.html | 4 | ||||
-rw-r--r-- | g4f/models.py | 302 |
2 files changed, 136 insertions, 170 deletions
diff --git a/g4f/gui/client/index.html b/g4f/gui/client/index.html index a2f883d9..1a660062 100644 --- a/g4f/gui/client/index.html +++ b/g4f/gui/client/index.html @@ -229,8 +229,8 @@ <option value="">Model: Default</option> <option value="gpt-4">gpt-4</option> <option value="gpt-3.5-turbo">gpt-3.5-turbo</option> - <option value="llama2-70b">llama2-70b</option> - <option value="llama3-70b-instruct">llama3-70b-instruct</option> + <option value="llama-3-70b-chat">llama-3-70b-chat</option> + <option value="llama-3.1-70b">llama-3.1-70b</option> <option value="gemini-pro">gemini-pro</option> <option value="">----</option> </select> diff --git a/g4f/models.py b/g4f/models.py index a8d9317a..bcc50b9f 100644 --- a/g4f/models.py +++ b/g4f/models.py @@ -6,38 +6,38 @@ from .Provider import IterListProvider, ProviderType from .Provider import ( AI365VIP, Allyfy, - Bing, - Blackbox, - ChatGot, - Chatgpt4o, - Chatgpt4Online, - ChatgptFree, - DDG, - DeepInfra, - DeepInfraImage, - FreeChatgpt, - FreeGpt, - Gemini, - GeminiPro, - GeminiProChat, - GigaChat, - HuggingChat, - HuggingFace, - Koala, - Liaobots, - Marsyoo, - MetaAI, - OpenaiChat, - PerplexityLabs, - Pi, - Pizzagpt, - Reka, - Replicate, - ReplicateHome, - Vercel, - You, + Bing, + Blackbox, + ChatGot, + Chatgpt4o, + Chatgpt4Online, + ChatgptFree, + DDG, + DeepInfra, + DeepInfraImage, + FreeChatgpt, + FreeGpt, + Gemini, + GeminiPro, + GeminiProChat, + GigaChat, + HuggingChat, + HuggingFace, + Koala, + Liaobots, + Marsyoo, + MetaAI, + OpenaiChat, + PerplexityLabs, + Pi, + Pizzagpt, + Reka, + Replicate, + ReplicateHome, + You, ) + @dataclass(unsafe_hash=True) class Model: """ @@ -198,40 +198,40 @@ meta = Model( best_provider = MetaAI ) -llama3_8b_instruct = Model( +llama_3_8b_instruct = Model( name = "meta-llama/Meta-Llama-3-8B-Instruct", base_provider = "meta", best_provider = IterListProvider([DeepInfra, PerplexityLabs, Replicate]) ) -llama3_70b_instruct = Model( +llama_3_70b_instruct = Model( name = "meta-llama/Meta-Llama-3-70B-Instruct", base_provider = "meta", - best_provider = IterListProvider([DeepInfra, PerplexityLabs, Replicate, DDG, ReplicateHome]) + best_provider = IterListProvider([DeepInfra, PerplexityLabs, Replicate]) ) -llama_3_1_70b_Instruct = Model( - name = "meta-llama/Meta-Llama-3.1-70B-Instruct", +llama3_70b_instruct = Model( + name = "meta/meta-llama-3-70b-instruct", base_provider = "meta", - best_provider = IterListProvider([HuggingChat, HuggingFace]) + best_provider = IterListProvider([ReplicateHome]) ) -llama_3_1_405b_Instruct_FP8 = Model( - name = "meta-llama/Meta-Llama-3.1-405B-Instruct-FP8", +llama_3_70b_chat_hf = Model( + name = "meta-llama/Llama-3-70b-chat-hf", base_provider = "meta", - best_provider = IterListProvider([HuggingChat, HuggingFace]) + best_provider = IterListProvider([DDG]) ) -codellama_34b_instruct = Model( - name = "codellama/CodeLlama-34b-Instruct-hf", +llama_3_1_70b_Instruct = Model( + name = "meta-llama/Meta-Llama-3.1-70B-Instruct", base_provider = "meta", - best_provider = HuggingChat + best_provider = IterListProvider([HuggingChat, HuggingFace]) ) -codellama_70b_instruct = Model( - name = "codellama/CodeLlama-70b-Instruct-hf", +llama_3_1_405b_Instruct_FP8 = Model( + name = "meta-llama/Meta-Llama-3.1-405B-Instruct-FP8", base_provider = "meta", - best_provider = IterListProvider([DeepInfra]) + best_provider = IterListProvider([HuggingChat, HuggingFace]) ) @@ -499,150 +499,116 @@ class ModelUtils: """ convert: dict[str, Model] = { -############ -### Text ### -############ - - ### OpenAI ### - ### GPT-3.5 / GPT-4 ### - # gpt-3.5 - 'gpt-3.5-turbo' : gpt_35_turbo, - 'gpt-3.5-turbo-0613' : gpt_35_turbo_0613, - 'gpt-3.5-turbo-16k' : gpt_35_turbo_16k, - 'gpt-3.5-turbo-16k-0613' : gpt_35_turbo_16k_0613, - 'gpt-3.5-long': gpt_35_long, - - # gpt-4 - 'gpt-4o' : gpt_4o, - 'gpt-4o-mini' : gpt_4o_mini, - 'gpt-4' : gpt_4, - 'gpt-4-0613' : gpt_4_0613, - 'gpt-4-32k' : gpt_4_32k, - 'gpt-4-32k-0613' : gpt_4_32k_0613, - 'gpt-4-turbo' : gpt_4_turbo, - - - ### Meta ### - "meta-ai": meta, - - 'llama3-8b': llama3_8b_instruct, # alias - 'llama3-70b': llama3_70b_instruct, # alias - 'llama3-8b-instruct' : llama3_8b_instruct, - 'llama3-70b-instruct': llama3_70b_instruct, - 'llama-3.1-70b-Instruct': llama_3_1_70b_Instruct, - 'llama-3.1-405B-Instruct-FP8': llama_3_1_405b_Instruct_FP8, + ############ + ### Text ### + ############ - - 'codellama-34b-instruct': codellama_34b_instruct, - 'codellama-70b-instruct': codellama_70b_instruct, - - - ### Mistral (Opensource) ### - 'mixtral-8x7b': mixtral_8x7b, - 'mistral-7b-v02': mistral_7b_v02, + ### OpenAI ### + ### GPT-3.5 / GPT-4 ### + # gpt-3.5 + 'gpt-3.5-turbo': gpt_35_turbo, + 'gpt-3.5-long': gpt_35_long, + + # gpt-4 + 'gpt-4o' : gpt_4o, + 'gpt-4o-mini' : gpt_4o_mini, + 'gpt-4' : gpt_4, + 'gpt-4-turbo' : gpt_4_turbo, + ### Meta ### + "meta-ai": meta, - ### NousResearch ### - 'Nous-Hermes-2-Mixtral-8x7B-DPO': Nous_Hermes_2_Mixtral_8x7B_DPO, - - - ### 01-ai ### - 'Yi-1.5-34B-Chat': Yi_1_5_34B_Chat, - - - ### Microsoft ### - 'Phi-3-mini-4k-instruct': Phi_3_mini_4k_instruct, - - - ### Google ### - # gemini - 'gemini': gemini, - 'gemini-pro': gemini_pro, - 'gemini-flash': gemini_flash, - - # gemma - 'gemma-2b-it': gemma_2b_it, - 'gemma-2-9b-it': gemma_2_9b_it, - 'gemma-2-27b-it': gemma_2_27b_it, - - - ### Anthropic ### - 'claude-2': claude_2, - 'claude-2.0': claude_2_0, - 'claude-2.1': claude_2_1, - 'claude-3-opus': claude_3_opus, - 'claude-3-sonnet': claude_3_sonnet, - 'claude-3-5-sonnet': claude_3_5_sonnet, - 'claude-3-haiku': claude_3_haiku, - - - ### Reka AI ### - 'reka': reka_core, - - - ### NVIDIA ### - 'nemotron-4-340b-instruct': nemotron_4_340b_instruct, - - - ### Blackbox ### - 'blackbox': blackbox, - - - ### CohereForAI ### - 'command-r+': command_r_plus, + 'llama-3-8b': llama_3_8b_instruct, + 'llama-3-70b': llama_3_70b_instruct, + 'llama-3-70b-chat': llama_3_70b_chat_hf, + 'llama-3-70b-instruct': llama3_70b_instruct, + 'llama-3.1-70b': llama_3_1_70b_Instruct, + 'llama-3.1-405b': llama_3_1_405b_Instruct_FP8, + ### Mistral (Opensource) ### + 'mixtral-8x7b': mixtral_8x7b, + 'mistral-7b-v02': mistral_7b_v02, - ### Databricks ### - 'dbrx-instruct': dbrx_instruct, - + ### NousResearch ### + 'Nous-Hermes-2-Mixtral-8x7B-DPO': Nous_Hermes_2_Mixtral_8x7B_DPO, - ### GigaChat ### - 'gigachat': gigachat, - - - ### iFlytek ### - 'SparkDesk-v1.1': SparkDesk_v1_1, + ### 01-ai ### + 'Yi-1.5-34b': Yi_1_5_34B_Chat, + ### Microsoft ### + 'Phi-3-mini-4k': Phi_3_mini_4k_instruct, + + ### Google ### + # gemini + 'gemini': gemini, + 'gemini-pro': gemini_pro, + 'gemini-flash': gemini_flash, - ### DeepSeek ### - 'deepseek-coder': deepseek_coder, - 'deepseek-chat': deepseek_chat, + # gemma + 'gemma-2b': gemma_2b_it, + 'gemma-2-9b': gemma_2_9b_it, + 'gemma-2-27b': gemma_2_27b_it, + + ### Anthropic ### + 'claude-2': claude_2, + 'claude-2.0': claude_2_0, + 'claude-2.1': claude_2_1, + 'claude-3-opus': claude_3_opus, + 'claude-3-sonnet': claude_3_sonnet, + 'claude-3-5-sonnet': claude_3_5_sonnet, + 'claude-3-haiku': claude_3_haiku, + + ### Reka AI ### + 'reka': reka_core, + + ### NVIDIA ### + 'nemotron-4-340b': nemotron_4_340b_instruct, + ### Blackbox ### + 'blackbox': blackbox, - ### ### Qwen ### ### - 'Qwen2-7B-Instruct': Qwen2_7B_Instruct, + ### CohereForAI ### + 'command-r+': command_r_plus, + ### Databricks ### + 'dbrx-instruct': dbrx_instruct, + + ### GigaChat ### + 'gigachat': gigachat, - ### Zhipu AI ### - 'glm4-9B-chat': glm4_9B_chat, - 'chatglm3-6B': chatglm3_6B, + ### iFlytek ### + 'SparkDesk-v1.1': SparkDesk_v1_1, + ### DeepSeek ### + 'deepseek-coder': deepseek_coder, + 'deepseek-chat': deepseek_chat, - ### 01-ai ### - 'Yi-1.5-9B-Chat': Yi_1_5_9B_Chat, + ### Qwen ### + 'Qwen2-7b': Qwen2_7B_Instruct, + ### Zhipu AI ### + 'glm4-9b': glm4_9B_chat, + 'chatglm3-6b': chatglm3_6B, - # Other - 'pi': pi, - + ### 01-ai ### + 'Yi-1.5-9b': Yi_1_5_9B_Chat, + # Other + 'pi': pi, -############# -### Image ### -############# - - ### Stability AI ### - 'sdxl': sdxl, - 'stable-diffusion-3': stable_diffusion_3, + ############# + ### Image ### + ############# - ### ByteDance ### - 'sdxl-lightning-4step': sdxl_lightning_4step, + ### Stability AI ### + 'sdxl': sdxl, + 'stable-diffusion-3': stable_diffusion_3, - ### ByteDance ### - 'sdxl-lightning-4step': sdxl_lightning_4step, + ### ByteDance ### + 'sdxl-lightning': sdxl_lightning_4step, - ### Playground ### - 'playground-v2.5-1024px-aesthetic': playground_v2_5_1024px_aesthetic, + ### Playground ### + 'playground-v2.5': playground_v2_5_1024px_aesthetic, } |