Add Phi3 preset

app.py

@@ -22,7 +22,7 @@ from contents import (
 from gradio_highlightedtextbox import HighlightedTextbox
 from gradio_modal import Modal
 from presets import (
-    set_chatml_preset,
+    set_phi3_preset,
     set_cora_preset,
     set_default_preset,
     set_mbart_mmt_preset,
@@ -244,10 +244,14 @@ with gr.Blocks(css=custom_css) as demo:
                     "Preset for the <a href='https://huggingface.co/gsarti/cora_mgen' target='_blank'>CORA Multilingual QA</a> model.\nUses special templates for inputs."
                 )
             with gr.Column():
-                chatml_template = gr.Button("Qwen ChatML", variant="secondary")
+                phi3_preset = gr.Button("Phi-3", variant="secondary", variant="secondary", interactive=False)
                 gr.Markdown(
-                    "Preset for models using the <a href='https://github.com/MicrosoftDocs/azure-docs/blob/main/articles/ai-services/openai/includes/chat-markup-language.md' target='_blank'>ChatML conversational template</a>.\nUses <code><|im_start|></code>, <code><|im_end|></code> special tokens."
+                    "Preset for the <a href='https://huggingface.co/phi3/phi3-chatbot' target='_blank'>Phi-3</a> conversational model.\nUses <code><|user|></code>, <code><|system|></code> and <code><|assistant|></code> and <code><|end|></code> special tokens."
                 )
+                #chatml_template = gr.Button("Qwen ChatML", variant="secondary")
+                #gr.Markdown(
+                #    "Preset for models using the <a href='https://github.com/MicrosoftDocs/azure-docs/blob/main/articles/ai-services/openai/includes/chat-markup-language.md' target='_blank'>ChatML conversational template</a>.\nUses <code><|im_start|></code>, <code><|im_end|></code> special tokens."
+                #)
         with gr.Row(equal_height=True):
             with gr.Column(scale=1):
                 mbart_mmt_template = gr.Button(
@@ -289,7 +293,7 @@ with gr.Blocks(css=custom_css) as demo:
                 )
                 gr.Markdown(
                     "Preset for models using the <a href='https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2' target='_blank'>Mistral Instruct template</a>.\nUses <code>[INST]...[/INST]</code> special tokens."
-                )         
+                )
         gr.Markdown("## ⚙️ PECoRe Parameters")
         with gr.Row(equal_height=True):
             with gr.Column():
@@ -592,8 +596,8 @@ with gr.Blocks(css=custom_css) as demo:
 
     check_enable_large_models.input(
         lambda checkbox, *buttons: [gr.Button(interactive=checkbox) for _ in buttons],
-        inputs=[check_enable_large_models, zephyr_preset, towerinstruct_template, gemma_template, mistral_instruct_template],
-        outputs=[zephyr_preset, towerinstruct_template, gemma_template, mistral_instruct_template],
+        inputs=[check_enable_large_models, phi3_preset, zephyr_preset, towerinstruct_template, gemma_template, mistral_instruct_template],
+        outputs=[phi3_preset, zephyr_preset, towerinstruct_template, gemma_template, mistral_instruct_template],
     )
 
     outputs_to_reset = [
@@ -646,11 +650,23 @@ with gr.Blocks(css=custom_css) as demo:
         outputs=[model_name_or_path, input_template, output_template, tokenizer_kwargs],
     ).success(preload_model, inputs=load_model_args, cancels=load_model_event)
 
-    chatml_template.click(**reset_kwargs).then(
-        set_chatml_preset,
+    # chatml_template.click(**reset_kwargs).then(
+    #     set_chatml_preset,
+    #     outputs=[
+    #         model_name_or_path,
+    #         input_template,
+    #         contextless_input_template,
+    #         special_tokens_to_keep,
+    #         generation_kwargs,
+    #     ],
+    # ).success(preload_model, inputs=load_model_args, cancels=load_model_event)
+
+    phi3_preset.click(**reset_kwargs).then(
+        set_phi3_preset,
         outputs=[
             model_name_or_path,
             input_template,
+            decoder_input_output_separator,
             contextless_input_template,
             special_tokens_to_keep,
             generation_kwargs,

presets.py

@@ -92,6 +92,16 @@ def set_mistral_instruct_preset():
         '{\n\t"max_new_tokens": 50\n}',  # generation_kwargs
     )
 
+def set_phi3_preset():
+    return (
+        "microsoft/Phi-3-mini-4k-instruct", # model_name_or_path
+        "<|system|>\n{system_prompt}<|end|>\n<|user|>\n{context}\n\n{current}<|end|>\n<|assistant|>".replace("{system_prompt}", SYSTEM_PROMPT), # input_template
+        "\n",  # decoder_input_output_separator
+        "<|system|>\n{system_prompt}<|end|>\n<|user|>\n{current}<|end|>\n<|assistant|>".replace("{system_prompt}", SYSTEM_PROMPT), # input_current_text_template
+        ["<|system|>", "<|end|>", "<|assistant|>", "<|user|>"], # special_tokens_to_keep
+        '{\n\t"max_new_tokens": 50\n}',  # generation_kwargs
+    )
+
 def update_code_snippets_fn(
     input_current_text: str,
     input_context_text: str,