Update space

apis/chat_api.py

@@ -61,6 +61,13 @@ class ChatAPIApp:
                     "created": 1700000000,
                     "owned_by": "NousResearch",
                 },
+                {
+                    "id": "falcon-40b-instruct-GPTQ-inference-endpoints",
+                    "description": "[philschmid/falcon-40b-instruct-GPTQ-inference-endpoints]: https://huggingface.co/philschmid/falcon-40b-instruct-GPTQ-inference-endpoints",
+                    "object": "model",
+                    "created": 1700000000,
+                    "owned_by": "philschmid",
+                }
             ],
         }
         return self.available_models

messagers/message_composer.py

@@ -9,6 +9,7 @@ class MessageComposer:
         "mixtral-8x7b",
         "mistral-7b",
         "openchat-3.5",
+        "falcon-40b-instruct-GPTQ-inference-endpoints",
         "nous-mixtral-8x7b",
     ]
 
@@ -116,6 +117,27 @@ class MessageComposer:
                     )
             self.merged_str_list.append(f"GPT4 Correct Assistant:\n")
             self.merged_str = "\n".join(self.merged_str_list)
+        elif self.model in ["falcon-40b-instruct-GPTQ-inference-endpoints"]:
+            self.messages = self.concat_messages_by_role(messages)
+            self.merged_str_list = []
+            self.end_of_turn = "<|end_of_turn|>"
+            for message in self.messages:
+                role = message["role"]
+                content = message["content"]
+                if role in self.inst_roles:
+                    self.merged_str_list.append(
+                        f"GPT4 Correct User:\n{content}{self.end_of_turn}"
+                    )
+                elif role in self.answer_roles:
+                    self.merged_str_list.append(
+                        f"GPT4 Correct Assistant:\n{content}{self.end_of_turn}"
+                    )
+                else:
+                    self.merged_str_list.append(
+                        f"GPT4 Correct User: {content}{self.end_of_turn}"
+                    )
+            self.merged_str_list.append(f"GPT4 Correct Assistant:\n")
+            self.merged_str = "\n".join(self.merged_str_list)
         else:
             self.merged_str = "\n".join(
                 [
@@ -206,6 +228,21 @@ class MessageComposer:
             self.append_last_instruction_to_messages(
                 inst_matches_list, pair_matches_list
             )
+        elif self.model in ["falcon-40b-instruct-GPTQ-inference-endpoints"]:
+            pair_pattern = r"GPT4 Correct User:(?P<inst>[\s\S]*?)<\|end_of_turn\|>\s*GPT4 Correct Assistant:(?P<answer>[\s\S]*?)<\|end_of_turn\|>"
+            pair_matches = re.finditer(
+                pair_pattern, self.merged_str, flags=re.MULTILINE | re.IGNORECASE
+            )
+            pair_matches_list = list(pair_matches)
+            self.messages = self.convert_pair_matches_to_messages(pair_matches_list)
+            inst_pattern = r"GPT4 Correct User:(?P<inst>[\s\S]*?)<\|end_of_turn\|>"
+            inst_matches = re.finditer(
+                inst_pattern, self.merged_str, flags=re.MULTILINE | re.IGNORECASE
+            )
+            inst_matches_list = list(inst_matches)
+            self.append_last_instruction_to_messages(
+                inst_matches_list, pair_matches_list
+            )
         else:
             self.messages = [
                 {

networks/message_streamer.py

@@ -13,6 +13,7 @@ class MessageStreamer:
         "mistral-7b": "mistralai/Mistral-7B-Instruct-v0.2",  # 65.71, fast
         "nous-mixtral-8x7b": "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO",
         "openchat-3.5": "openchat/openchat-3.5-0106",  # 68.89, fast
+        "falcon-40b-instruct-GPTQ-inference-endpoints": "philschmid/falcon-40b-instruct-GPTQ-inference-endpoints",
         # "zephyr-7b-beta": "HuggingFaceH4/zephyr-7b-beta",  # ❌ Too Slow
         # "llama-70b": "meta-llama/Llama-2-70b-chat-hf",  # ❌ Require Pro User
         # "codellama-34b": "codellama/CodeLlama-34b-Instruct-hf",  # ❌ Low Score
@@ -24,12 +25,16 @@ class MessageStreamer:
         "mistral-7b": "</s>",
         "nous-mixtral-8x7b": "<|im_end|>",
         "openchat-3.5": "<|end_of_turn|>",
+        "falcon-40b-instruct-GPTQ-inference-endpoints": "<|end_of_turn|>",
+
     }
     TOKEN_LIMIT_MAP = {
         "mixtral-8x7b": 32768,
         "mistral-7b": 32768,
         "nous-mixtral-8x7b": 32768,
         "openchat-3.5": 8192,
+        "falcon-40b-instruct-GPTQ-inference-endpoints": 8192,
+
     }
     TOKEN_RESERVED = 100