From 5fc16a2474bbef31b8286929f66f584dd5824316 Mon Sep 17 00:00:00 2001 From: Cohee <18619528+Cohee1207@users.noreply.github.com> Date: Mon, 26 Aug 2024 12:07:36 +0300 Subject: [PATCH] New AI21 Jamba + tokenizer --- default/content/presets/openai/Default.json | 5 +- default/content/settings.json | 6 +- public/index.html | 46 ++------ public/script.js | 2 +- public/scripts/openai.js | 113 +++----------------- public/scripts/tokenizers.js | 49 +++++---- src/endpoints/backends/chat-completions.js | 85 ++++++--------- src/endpoints/tokenizers.js | 74 +++++-------- src/prompt-converters.js | 74 +++++++++++++ src/tokenizers/jamba.model | Bin 0 -> 1124714 bytes 10 files changed, 188 insertions(+), 266 deletions(-) create mode 100644 src/tokenizers/jamba.model diff --git a/default/content/presets/openai/Default.json b/default/content/presets/openai/Default.json index e2345c687..81e82a389 100644 --- a/default/content/presets/openai/Default.json +++ b/default/content/presets/openai/Default.json @@ -8,7 +8,7 @@ "openrouter_force_instruct": false, "openrouter_group_models": false, "openrouter_sort_models": "alphabetically", - "ai21_model": "j2-ultra", + "ai21_model": "jamba-1.5-mini", "mistralai_model": "mistral-large-latest", "custom_model": "", "custom_url": "", @@ -19,7 +19,6 @@ "temperature": 1, "frequency_penalty": 0, "presence_penalty": 0, - "count_penalty": 0, "top_p": 1, "top_k": 0, "top_a": 0, @@ -233,8 +232,6 @@ "assistant_prefill": "", "assistant_impersonation": "", "human_sysprompt_message": "Let's get started. Please generate your response based on the information and instructions provided above.", - "use_ai21_tokenizer": false, - "use_google_tokenizer": false, "claude_use_sysprompt": false, "use_alt_scale": false, "squash_system_messages": false, diff --git a/default/content/settings.json b/default/content/settings.json index a8af5277c..4bb36d9ea 100644 --- a/default/content/settings.json +++ b/default/content/settings.json @@ -452,7 +452,6 @@ "temp_openai": 1.0, "freq_pen_openai": 0, "pres_pen_openai": 0, - "count_pen": 0, "top_p_openai": 1, "top_k_openai": 0, "stream_openai": true, @@ -614,7 +613,7 @@ "wi_format": "{0}", "openai_model": "gpt-4-turbo", "claude_model": "claude-3-5-sonnet-20240620", - "ai21_model": "j2-ultra", + "ai21_model": "jamba-1.5-mini", "windowai_model": "", "openrouter_model": "OR_Website", "jailbreak_system": true, @@ -625,7 +624,6 @@ "show_external_models": false, "proxy_password": "", "assistant_prefill": "", - "assistant_impersonation": "", - "use_ai21_tokenizer": false + "assistant_impersonation": "" } } diff --git a/public/index.html b/public/index.html index 46962677b..a411bf235 100644 --- a/public/index.html +++ b/public/index.html @@ -436,7 +436,7 @@ -
+
Frequency Penalty
@@ -449,7 +449,7 @@
-
+
Presence Penalty
@@ -462,20 +462,7 @@
-
-
- Count Penalty -
-
-
- -
-
- -
-
-
-
+
Top K
@@ -1791,22 +1778,6 @@
-
- -
- Use the appropriate tokenizer for Jurassic models, which is more efficient than GPT's. -
-
-
- -
- Use the appropriate tokenizer for Google models via their API. Slower prompt processing, but offers much more accurate token counting. -
-