crfm-helm 0.2.1__py3-none-any.whl → 0.2.2__py3-none-any.whl

helm/proxy/models.py

@@ -11,8 +11,14 @@ EMBEDDING_MODEL_TAG: str = "embedding"
 FULL_FUNCTIONALITY_TEXT_MODEL_TAG: str = "full_functionality_text"
 LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG: str = "limited_functionality_text"
 
+# ChatML format
+CHATML_MODEL_TAG: str = "chatml"
+
 # For OpenAI models with wider context windows
-WIDER_CONTEXT_WINDOW_TAG: str = "wider_context_window"
+WIDER_CONTEXT_WINDOW_TAG: str = "wider_context_window"  # 4000 tokens
+
+# For AI21 Jurassic-2 models with wider context windows
+AI21_WIDER_CONTEXT_WINDOW_TAG: str = "ai21_wider_context_window"
 
 # To fetch models that use these tokenizers
 GPT2_TOKENIZER_TAG: str = "gpt2_tokenizer"
@@ -122,6 +128,31 @@ ALL_MODELS = [
         description="Jurassic-1 Large (7.5B parameters)",
         tags=[TEXT_MODEL_TAG, FULL_FUNCTIONALITY_TEXT_MODEL_TAG, AI21_TOKENIZER_TAG],
     ),
+    # AI21 Jurassic-2 Models: https://www.ai21.com/blog/introducing-j2
+    Model(
+        group="jurassic",
+        creator_organization="AI21 Labs",
+        name="ai21/j2-jumbo",
+        display_name="Jurassic-2 Jumbo (178B)",
+        description="Jurassic-2 Jumbo (178B parameters)",
+        tags=[TEXT_MODEL_TAG, FULL_FUNCTIONALITY_TEXT_MODEL_TAG, AI21_TOKENIZER_TAG],
+    ),
+    Model(
+        group="jurassic",
+        creator_organization="AI21 Labs",
+        name="ai21/j2-grande",
+        display_name="Jurassic-2 Grande (17B)",
+        description="Jurassic-2 Grande (17B parameters) with a few tweaks to the training process.",
+        tags=[TEXT_MODEL_TAG, AI21_WIDER_CONTEXT_WINDOW_TAG, FULL_FUNCTIONALITY_TEXT_MODEL_TAG, AI21_TOKENIZER_TAG],
+    ),
+    Model(
+        group="jurassic",
+        creator_organization="AI21 Labs",
+        name="ai21/j2-large",
+        display_name="Jurassic-2 Large (7.5B)",
+        description="Jurassic-2 Large (7.5B parameters)",
+        tags=[TEXT_MODEL_TAG, AI21_WIDER_CONTEXT_WINDOW_TAG, FULL_FUNCTIONALITY_TEXT_MODEL_TAG, AI21_TOKENIZER_TAG],
+    ),
     # Aleph Alpha's Luminous models: https://docs.aleph-alpha.com/docs/introduction/luminous
     Model(
         group="luminous",
@@ -250,6 +281,24 @@ ALL_MODELS = [
         description="Cohere small v20220720 (410M parameters)",
         tags=[TEXT_MODEL_TAG, FULL_FUNCTIONALITY_TEXT_MODEL_TAG, COHERE_TOKENIZER_TAG],
     ),
+    Model(
+        group="cohere",
+        creator_organization="Cohere",
+        name="cohere/command-medium-beta",
+        display_name="Cohere Command beta (6.1B)",
+        description="Cohere Command beta (6.1B parameters) is fine-tuned from the medium model "
+        "to respond well with instruction-like prompts",
+        tags=[TEXT_MODEL_TAG, FULL_FUNCTIONALITY_TEXT_MODEL_TAG, COHERE_TOKENIZER_TAG],
+    ),
+    Model(
+        group="cohere",
+        creator_organization="Cohere",
+        name="cohere/command-xlarge-beta",
+        display_name="Cohere Command beta (52.4B)",
+        description="Cohere Command beta (52.4B parameters) is fine-tuned from the XL model "
+        "to respond well with instruction-like prompts",
+        tags=[TEXT_MODEL_TAG, FULL_FUNCTIONALITY_TEXT_MODEL_TAG, COHERE_TOKENIZER_TAG],
+    ),
     # EleutherAI
     Model(
         group="together",
@@ -323,6 +372,15 @@ ALL_MODELS = [
         # Does not support echo=True
         tags=[TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, ABLATION_MODEL_TAG, NO_NEWLINES_TAG],
     ),
+    Model(
+        group="together",
+        creator_organization="Google",
+        name="together/flan-t5-xxl",
+        display_name="Flan-T5 (11B)",
+        description="Flan-T5 (11B parameters) is T5 fine-tuned on 1.8K tasks.",
+        # Does not support echo=True
+        tags=[TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, ABLATION_MODEL_TAG, NO_NEWLINES_TAG],
+    ),
     Model(
         group="together",
         creator_organization="Google",
@@ -498,7 +556,21 @@ ALL_MODELS = [
         description="Code model that is a stronger, multilingual version of the Codex (12B) model in the paper.",
         tags=[CODE_MODEL_TAG, GPT2_TOKENIZER_TAG],
     ),
-    # ChatGPT - https://openai.com/blog/chatgpt
+    # ChatGPT: https://openai.com/blog/chatgpt
+    Model(
+        group="gpt3",
+        creator_organization="OpenAI",
+        name="openai/gpt-3.5-turbo-0301",
+        display_name="gpt-3.5-turbo-0301",
+        # https://platform.openai.com/docs/models/gpt-3-5
+        description="Sibling model of text-davinci-003 is optimized for chat but works well "
+        "for traditional completions tasks as well. Snapshot from 2023-03-01.",
+        # The claimed sequence length is 4096, but as of 2023-03-07, the empirical usable
+        # sequence length is smaller at 4087 with one user input message and one assistant
+        # output message because ChatGPT uses special tokens for message roles and boundaries.
+        # We use a rounded-down sequence length of 4000 to account for these special tokens.
+        tags=[TEXT_MODEL_TAG, WIDER_CONTEXT_WINDOW_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, GPT2_TOKENIZER_TAG],
+    ),
     Model(
         group="gpt3",
         creator_organization="OpenAI",
@@ -550,6 +622,14 @@ ALL_MODELS = [
         description="GPT-JT (6B parameters) is a fork of GPT-J",
         tags=[TEXT_MODEL_TAG, FULL_FUNCTIONALITY_TEXT_MODEL_TAG, GPTJ_TOKENIZER_TAG],
     ),
+    Model(
+        group="together",
+        creator_organization="Together",
+        name="together/gpt-neoxt-chat-base-20b",
+        display_name="GPT-NeoXT-Chat-Base (20B)",
+        description="GPT-NeoXT-Chat-Base (20B parameters) is a fork of GPT-NeoX",
+        tags=[TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, CHATML_MODEL_TAG, GPTNEO_TOKENIZER_TAG],
+    ),
     # Tsinghua
     Model(
         group="together",