Updated certainty and answerability models (#9)

- Updated loras for answerability and citations (b3b1d010db86154c687b670c7ed94243af2f44eb)


Co-authored-by: Fred Reiss <frreiss@users.noreply.huggingface.co>

answerability/lora/gpt-oss-20b/adapter_config.json

@@ -0,0 +1,45 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "openai/gpt-oss-20b",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 32,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": [
+    "7.mlp.experts.gate_up_proj",
+    "7.mlp.experts.down_proj",
+    "15.mlp.experts.gate_up_proj",
+    "15.mlp.experts.down_proj",
+    "23.mlp.experts.gate_up_proj",
+    "23.mlp.experts.down_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

answerability/lora/gpt-oss-20b/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ee17447d06e0327bed4cd6811da1fa2607520285bb35c712bccb8d6d7f9e772
+size 219238968

answerability/lora/gpt-oss-20b/io.yaml

@@ -0,0 +1,19 @@
+# Model name string, or null to use whatever is provided in the chat completion request
+model: ~
+# JSON schema of the model's output
+response_format: |
+  {
+    "type": "string",
+    "enum": ["answerable", "unanswerable"]
+  }
+transformations:
+  - type: likelihood
+    categories_to_values:
+      "answerable": 1.0
+      "unanswerable": 0.0
+    input_path: []
+    output_name: ~  # Null == no change
+instruction: ~
+parameters:
+  max_completion_tokens: 6
+sentence_boundaries: false

citations/lora/granite-3.3-2b-instruct/adapter_config.json

@@ -3,9 +3,6 @@
   "auto_mapping": null,
   "base_model_name_or_path": "ibm-granite/granite-3.3-2b-instruct",
   "bias": "none",
-  "corda_config": null,
-  "eva_config": null,
-  "exclude_modules": null,
   "fan_in_fan_out": false,
   "inference_mode": true,
   "init_lora_weights": true,
@@ -14,8 +11,7 @@
   "layers_to_transform": null,
   "loftq_config": {},
   "lora_alpha": 32,
-  "lora_bias": false,
-  "lora_dropout": 0.05,
+  "lora_dropout": 0.1,
   "megatron_config": null,
   "megatron_core": "megatron.core",
   "modules_to_save": null,
@@ -24,12 +20,15 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "k_proj",
     "q_proj",
-    "v_proj"
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "up_proj",
+    "down_proj",
+    "gate_proj"
   ],
   "task_type": "CAUSAL_LM",
-  "trainable_token_indices": null,
   "use_dora": false,
   "use_rslora": false
 }

citations/lora/granite-3.3-2b-instruct/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6957e303240dc287b6d894898fde78736f15d5057fd8f7b8551cf3c92e8ff4e2
-size 23625040
+oid sha256:be6dc64b4c7737c74a4c37efca2dc621cd8ac5f918be723aec23f59bf7873c2c
+size 56435848

citations/lora/granite-3.3-2b-instruct/io.yaml

@@ -0,0 +1,96 @@
+# Model name string, or null to use whatever is provided in the chat completion request
+model: ~
+# JSON schema of the model's output
+response_format: |
+  {
+    "$defs": {
+      "_MODEL_OUTPUT_ENTRY": {
+        "properties": {
+          "r": {
+            "minimum": 0,
+            "title": "R",
+            "type": "integer"
+          },
+          "c": {
+            "items": {
+              "minimum": 0,
+              "type": "integer"
+            },
+            "title": "C",
+            "type": "array"
+          }
+        },
+        "required": [
+          "r",
+          "c"
+        ],
+        "title": "_MODEL_OUTPUT_ENTRY",
+        "type": "object"
+      }
+    },
+    "items": {
+      "$ref": "#/$defs/_MODEL_OUTPUT_ENTRY"
+    },
+    "title": "_MODEL_OUTPUT",
+    "type": "array"
+  }
+transformations:
+  # Explode the list of document sentences in each citation
+  - type: explode
+    input_path: []  # Zero-length path means match root element
+    target_field: "c"
+  # Model may repeat itself; drop the resulting duplicates.
+  - type: drop_duplicates
+    input_path: []  # Zero-length path means match root element
+    target_fields: ["r", "c"]
+  # Replace sentence number with sentence location and contents.
+  # Do this first for sentences from the last turn, then for sentences from documents.
+  - type: decode_sentences
+    source: "last_message"
+    input_path: [~, "r"]  # Null in path means wildcard
+    # New fields to add for each sentence
+    output_names:
+      begin: "response_begin"
+      end: "response_end"
+      text: "response_text"
+  - type: decode_sentences
+    source: "documents"
+    input_path: [~, "c"]  # Null in path means wildcard
+    # New fields to add for each sentence
+    output_names:
+      document_id: "citation_doc_id"
+      begin: "citation_begin"
+      end: "citation_end"
+      text: "citation_text"
+  # Remove fields that we no longer need
+  - type: project
+    input_path: []
+    retained_fields:
+    - "response_begin"
+    - "response_end"
+    - "response_text"
+    - "citation_doc_id"
+    - "citation_begin"
+    - "citation_end"
+    - "citation_text"
+  # Merge adjacent document spans
+  - type: merge_spans
+    input_path: []
+    group_fields: ["response_begin", "response_end", "response_text", "citation_doc_id"]
+    begin_field: "citation_begin"
+    end_field: "citation_end"
+    text_field: "citation_text"
+
+instruction: >
+  Split the last assistant response into individual sentences. 
+  For each sentence in the response, identify the statement IDs from the below 
+  documents that it references. Ensure that your output includes all response 
+  sentence IDs, and for each response sentence ID, provide the list of corresponding 
+  referring document sentence IDs. The output must be a json structure.
+parameters:
+  max_completion_tokens: 4096
+sentence_boundaries:
+  # Mapping from string location to sentence delimiter prefix
+  last_message: "r"  # <r0>, <r1>, etc.
+  documents: "c"
+  

citations/lora/granite-3.3-2b-instruct/tokenizer_config.json

@@ -221,12 +221,11 @@
     "<|end_of_plugin|>"
   ],
   "bos_token": "<|end_of_text|>",
-  "chat_template": "{# Alias tools -> available_tools #}\n{%- if tools and not available_tools -%}\n    {%- set available_tools = tools -%}\n{%- endif -%}\n{%- if messages[0]['role'] == 'system' %}\n     {%- set system_message = messages[0]['content'] %}\n     {%- set loop_messages = messages[1:] %}\n {%- else %}\n     {%- set system_message = \"Knowledge Cutoff Date: April 2024.\nToday's Date: \" + strftime_now('%B %d, %Y') + \".\nYou are Granite, developed by IBM.\" %}\n     {%- if available_tools and documents %}\n         {%- set system_message = system_message + \" You are a helpful assistant with access to the following tools. When a tool is required to answer the user's query, respond only with <|tool_call|> followed by a JSON list of tools used. If a tool does not exist in the provided list of tools, notify the user that you do not have the ability to fulfill the request.\nWrite the response to the user's input by strictly aligning with the facts in the provided documents. If the information needed to answer the question is not available in the documents, inform the user that the question cannot be answered based on the available data.\" %}\n     {%- elif available_tools %}\n         {%- set system_message = system_message + \" You are a helpful assistant with access to the following tools. When a tool is required to answer the user's query, respond only with <|tool_call|> followed by a JSON list of tools used. If a tool does not exist in the provided list of tools, notify the user that you do not have the ability to fulfill the request.\" %}\n     {%- elif documents %}\n         {%- set system_message = system_message + \" Write the response to the user's input by strictly aligning with the facts in the provided documents. If the information needed to answer the question is not available in the documents, inform the user that the question cannot be answered based on the available data.\" %}\n    {%- elif thinking %}\n    {%- set system_message = system_message + \" You are a helpful AI assistant.\nRespond to every user query in a comprehensive and detailed way. You can write down your thoughts and reasoning process before responding. In the thought process, engage in a comprehensive cycle of analysis, summarization, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. In the response section, based on various attempts, explorations, and reflections from the thoughts section, systematically present the final solution that you deem correct. The response should summarize the thought process. Write your thoughts between <think></think> and write your response between <response></response> for each user query.\" %}\n     {%- else %}\n         {%- set system_message = system_message + \" You are a helpful AI assistant.\" %}\n     {%- endif %}\n     {%- if 'citations' in controls and documents %}\n         {%- set system_message = system_message + '\nUse the symbols <|start_of_cite|> and <|end_of_cite|> to indicate when a fact comes from a document in the search result, e.g <|start_of_cite|> {document_id: 1}my fact <|end_of_cite|> for a fact from document 1. Afterwards, list all the citations with their corresponding documents in an ordered list.' %}\n     {%- endif %}\n     {%- if 'hallucinations' in controls and documents %}\n         {%- set system_message = system_message + '\nFinally, after the response is written, include a numbered list of sentences from the response with a corresponding risk value that are hallucinated and not based in the documents.' %}\n     {%- endif %}\n     {%- set loop_messages = messages %}\n {%- endif %}\n {{- '<|start_of_role|>system<|end_of_role|>' + system_message + '<|end_of_text|>\n' }}\n {%- if available_tools %}\n     {{- '<|start_of_role|>available_tools<|end_of_role|>' }}\n     {{- available_tools | tojson(indent=4) }}\n     {{- '<|end_of_text|>\n' }}\n {%- endif %}\n {%- if documents %}\n     {%- for document in documents %}\n         {{- '<|start_of_role|>document {\"document_id\": \"' + document['doc_id'] | string + '\"}<|end_of_role|>\n' }}\n         {{- document['text'] }}\n         {{- '<|end_of_text|>\n' }}\n              {%- endfor %}\n {%- endif %}\n {%- for message in loop_messages %}\n     {{- '<|start_of_role|>' + message['role'] + '<|end_of_role|>' + message['content'] + '<|end_of_text|>\n' }}\n     {%- if loop.last and add_generation_prompt %}\n         {{- '<|start_of_role|>assistant' }}\n             {%- if controls %}\n                 {{- ' ' + controls | tojson()}}\n             {%- endif %}\n         {{- '<|end_of_role|>' }}\n     {%- endif %}\n {%- endfor %}",
+  "chat_template": "\n{# Alias tools -> available_tools #}\n{%- if tools and not available_tools -%}\n    {%- set available_tools = tools -%}\n{%- endif -%}\n{%- if messages[0]['role'] == 'system' %}\n     {%- set system_message = messages[0]['content'] %}\n     {%- set loop_messages = messages[1:] %}\n {%- else %}\n     {%- set system_message = \"Knowledge Cutoff Date: April 2024.\nToday's Date: \" + strftime_now('%B %d, %Y') + \".\nYou are Granite, developed by IBM.\" %}\n     {%- if available_tools and documents %}\n         {%- set system_message = system_message + \" You are a helpful assistant with access to the following tools. When a tool is required to answer the user's query, respond only with <|tool_call|> followed by a JSON list of tools used. If a tool does not exist in the provided list of tools, notify the user that you do not have the ability to fulfill the request.\nWrite the response to the user's input by strictly aligning with the facts in the provided documents. If the information needed to answer the question is not available in the documents, inform the user that the question cannot be answered based on the available data.\" %}\n     {%- elif available_tools %}\n         {%- set system_message = system_message + \" You are a helpful assistant with access to the following tools. When a tool is required to answer the user's query, respond only with <|tool_call|> followed by a JSON list of tools used. If a tool does not exist in the provided list of tools, notify the user that you do not have the ability to fulfill the request.\" %}\n     {%- elif documents %}\n         {%- set system_message = system_message + \" Write the response to the user's input by strictly aligning with the facts in the provided documents. If the information needed to answer the question is not available in the documents, inform the user that the question cannot be answered based on the available data.\" %}\n    {%- elif thinking %}\n    {%- set system_message = system_message + \" You are a helpful AI assistant.\nRespond to every user query in a comprehensive and detailed way. You can write down your thoughts and reasoning process before responding. In the thought process, engage in a comprehensive cycle of analysis, summarization, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. In the response section, based on various attempts, explorations, and reflections from the thoughts section, systematically present the final solution that you deem correct. The response should summarize the thought process. Write your thoughts between <think></think> and write your response between <response></response> for each user query.\" %}\n     {%- else %}\n         {%- set system_message = system_message + \" You are a helpful AI assistant.\" %}\n     {%- endif %}\n     {%- if 'citations' in controls and documents %}\n         {%- set system_message = system_message + '\nUse the symbols <|start_of_cite|> and <|end_of_cite|> to indicate when a fact comes from a document in the search result, e.g <|start_of_cite|> {document_id: 1}my fact <|end_of_cite|> for a fact from document 1. Afterwards, list all the citations with their corresponding documents in an ordered list.' %}\n     {%- endif %}\n     {%- if 'hallucinations' in controls and documents %}\n         {%- set system_message = system_message + '\nFinally, after the response is written, include a numbered list of sentences from the response with a corresponding risk value that are hallucinated and not based in the documents.' %}\n     {%- endif %}\n     {%- set loop_messages = messages %}\n {%- endif %}\n {{- '<|start_of_role|>system<|end_of_role|>' + system_message + '<|end_of_text|>\n' }}\n {%- if available_tools %}\n     {{- '<|start_of_role|>available_tools<|end_of_role|>' }}\n     {{- available_tools | tojson(indent=4) }}\n     {{- '<|end_of_text|>\n' }}\n {%- endif %}\n {%- if documents %}\n     {%- for document in documents %}\n         {{- '<|start_of_role|>document {\"document_id\": \"' + document['doc_id'] | string + '\"}<|end_of_role|>\n' }}\n         {{- document['text'] }}\n         {{- '<|end_of_text|>\n' }}\n              {%- endfor %}\n {%- endif %}\n {%- for message in loop_messages %}\n     {{- '<|start_of_role|>' + message['role'] + '<|end_of_role|>' + message['content'] + '<|end_of_text|>\n' }}\n     {%- if loop.last and add_generation_prompt %}\n         {{- '<|start_of_role|>assistant' }}\n             {%- if controls %}\n                 {{- ' ' + controls | tojson()}}\n             {%- endif %}\n         {{- '<|end_of_role|>' }}\n     {%- endif %}\n {%- endfor %}\n",
   "clean_up_tokenization_spaces": true,
   "eos_token": "<|end_of_text|>",
   "errors": "replace",
   "extra_special_tokens": {},
-  "legacy": true,
   "model_max_length": 9223372036854775807,
   "pad_token": "<|end_of_text|>",
   "padding_side": "left",

citations/lora/granite-3.3-8b-instruct/io.yaml

@@ -0,0 +1,96 @@
+# Model name string, or null to use whatever is provided in the chat completion request
+model: ~
+# JSON schema of the model's output
+response_format: |
+  {
+    "$defs": {
+      "_MODEL_OUTPUT_ENTRY": {
+        "properties": {
+          "r": {
+            "minimum": 0,
+            "title": "R",
+            "type": "integer"
+          },
+          "c": {
+            "items": {
+              "minimum": 0,
+              "type": "integer"
+            },
+            "title": "C",
+            "type": "array"
+          }
+        },
+        "required": [
+          "r",
+          "c"
+        ],
+        "title": "_MODEL_OUTPUT_ENTRY",
+        "type": "object"
+      }
+    },
+    "items": {
+      "$ref": "#/$defs/_MODEL_OUTPUT_ENTRY"
+    },
+    "title": "_MODEL_OUTPUT",
+    "type": "array"
+  }
+transformations:
+  # Explode the list of document sentences in each citation
+  - type: explode
+    input_path: []  # Zero-length path means match root element
+    target_field: "c"
+  # Model may repeat itself; drop the resulting duplicates.
+  - type: drop_duplicates
+    input_path: []  # Zero-length path means match root element
+    target_fields: ["r", "c"]
+  # Replace sentence number with sentence location and contents.
+  # Do this first for sentences from the last turn, then for sentences from documents.
+  - type: decode_sentences
+    source: "last_message"
+    input_path: [~, "r"]  # Null in path means wildcard
+    # New fields to add for each sentence
+    output_names:
+      begin: "response_begin"
+      end: "response_end"
+      text: "response_text"
+  - type: decode_sentences
+    source: "documents"
+    input_path: [~, "c"]  # Null in path means wildcard
+    # New fields to add for each sentence
+    output_names:
+      document_id: "citation_doc_id"
+      begin: "citation_begin"
+      end: "citation_end"
+      text: "citation_text"
+  # Remove fields that we no longer need
+  - type: project
+    input_path: []
+    retained_fields:
+    - "response_begin"
+    - "response_end"
+    - "response_text"
+    - "citation_doc_id"
+    - "citation_begin"
+    - "citation_end"
+    - "citation_text"
+  # Merge adjacent document spans
+  - type: merge_spans
+    input_path: []
+    group_fields: ["response_begin", "response_end", "response_text", "citation_doc_id"]
+    begin_field: "citation_begin"
+    end_field: "citation_end"
+    text_field: "citation_text"
+
+instruction: >
+  Split the last assistant response into individual sentences. 
+  For each sentence in the response, identify the statement IDs from the below 
+  documents that it references. Ensure that your output includes all response 
+  sentence IDs, and for each response sentence ID, provide the list of corresponding 
+  referring document sentence IDs. The output must be a json structure.
+parameters:
+  max_completion_tokens: 4096
+sentence_boundaries:
+  # Mapping from string location to sentence delimiter prefix
+  last_message: "r"  # <r0>, <r1>, etc.
+  documents: "c"
+