Merge pull request #13 from chtanch/qwen-template

Add instruction template for Qwen and chatGLM3

instruction-templates/ChatGLM3.yaml

@@ -0,0 +1,24 @@
+instruction_template: |-
+  {%- set ns = namespace(found=false) -%}
+  {%- for message in messages -%}
+      {%- if message['role'] == 'system' -%}
+          {%- set ns.found = true -%}
+      {%- endif -%}
+  {%- endfor -%}
+  {%- if not ns.found -%}
+      {{- '' + '' + '' -}}
+  {%- endif %}
+  {%- for message in messages %}
+      {%- if message['role'] == 'system' -%}
+          {{- '<|system|>\n' + message['content'] + '\n' -}}
+      {%- else -%}
+          {%- if message['role'] == 'user' -%}
+              {{-'<|user|>\n' + message['content'] + '\n'-}}
+          {%- else -%}
+              {{-'<|assistant|>\n' + message['content'] + '\n' -}}
+          {%- endif -%}
+      {%- endif -%}
+  {%- endfor -%}
+  {%- if add_generation_prompt -%}
+      {{-'<|assistant|>\n'-}}
+  {%- endif -%}

instruction-templates/Qwen.yaml

@@ -0,0 +1,24 @@
+instruction_template: |-
+  {%- set ns = namespace(found=false) -%}
+  {%- for message in messages -%}
+      {%- if message['role'] == 'system' -%}
+          {%- set ns.found = true -%}
+      {%- endif -%}
+  {%- endfor -%}
+  {%- if not ns.found -%}
+      {{- '' + '' + '' -}}
+  {%- endif %}
+  {%- for message in messages %}
+      {%- if message['role'] == 'system' -%}
+          {{- '<|im_start|>system\n' + message['content'] + '<|im_end|>\n' -}}
+      {%- else -%}
+          {%- if message['role'] == 'user' -%}
+              {{-'<|im_start|>user\n' + message['content'] + '<|im_end|>\n'-}}
+          {%- else -%}
+              {{-'<|im_start|>assistant\n' + message['content'] + '<|im_end|>\n' -}}
+          {%- endif -%}
+      {%- endif -%}
+  {%- endfor -%}
+  {%- if add_generation_prompt -%}
+      {{-'<|im_start|>assistant\n'-}}
+  {%- endif -%}

models/config.yaml

@@ -190,3 +190,5 @@
   instruction_template: 'ChatML'
 .*synthia:
   instruction_template: 'Synthia'
+.*qwen:
+  instruction_template: 'Qwen'

modules/text_generation.py

@@ -314,17 +314,17 @@ def generate_reply_HF(question, original_question, seed, state, stopping_strings
         generate_params.update({'synced_gpus': True})
 
     #tune the prompt based on qwen
-    QWEN_PROMPT_FORMAT = """
-    <|im_start|>system
-    You are a helpful assistant.
-    <|im_end|>
-    <|im_start|>user
-    {prompt}
-    <|im_end|>
-    <|im_start|>assistant
-    """
-    if shared.model.config.model_type == "qwen":
-        question = QWEN_PROMPT_FORMAT.format(prompt=question)
+    # QWEN_PROMPT_FORMAT = """
+    # <|im_start|>system
+    # You are a helpful assistant.
+    # <|im_end|>
+    # <|im_start|>user
+    # {prompt}
+    # <|im_end|>
+    # <|im_start|>assistant
+    # """
+    # if shared.model.config.model_type == "qwen":
+    #     question = QWEN_PROMPT_FORMAT.format(prompt=question)
 
     # Encode the input
     input_ids = encode(question, add_bos_token=state['add_bos_token'], truncation_length=get_max_prompt_length(state))