Merge pull request #219 from valory-xyz/chore/fix-llm-provider

auto-infer llm provider

packages/napthaai/customs/prediction_request_rag/component.yaml

@@ -7,12 +7,12 @@ license: Apache-2.0
 aea_version: '>=1.0.0, <2.0.0'
 fingerprint:
   __init__.py: bafybeibt7f7crtwvmkg7spy3jhscmlqltvyblzp32g6gj44v7tlo5lycuq
-  prediction_request_rag.py: bafybeibtvuddvbhjlyd4sbk7rwz4mcsr4hiigfgrpdhzwa6vn6bhb6fboy
+  prediction_request_rag.py: bafybeicllugnruskdj7ipmrj2vrtlxmjpqtwlk4c3cfjttfzuvkeldp3m4
 fingerprint_ignore_patterns: []
 entry_point: prediction_request_rag.py
 callable: run
 params:
-  default_model: claude-3-sonnet-20240229
+  default_model: gpt-4-0125-preview
 dependencies:
   google-api-python-client:
     version: ==2.95.0

packages/napthaai/customs/prediction_request_rag/prediction_request_rag.py

@@ -40,11 +40,16 @@ class LLMClientManager:
     """Client context manager for LLMs."""
 
     def __init__(
-        self, api_keys: List, llm_provider: str = None, embedding_provider: str = None
+        self, api_keys: List, model: str = None, embedding_provider: str = None
     ):
         self.api_keys = api_keys
-        self.llm_provider = llm_provider
         self.embedding_provider = embedding_provider
+        if "gpt" in model:
+            self.llm_provider = "openai"
+        elif "claude" in model:
+            self.llm_provider = "anthropic"
+        else:
+            self.llm_provider = "openrouter"
 
     def __enter__(self):
         clients = []
@@ -220,6 +225,9 @@ def embeddings(self, model, input):
 }
 ALLOWED_TOOLS = [
     "prediction-request-rag",
+
+    # LEGACY
+    "prediction-request-rag-claude",
 ]
 ALLOWED_MODELS = list(LLM_SETTINGS.keys())
 DEFAULT_NUM_URLS = defaultdict(lambda: 3)
@@ -300,8 +308,8 @@ def multi_queries(
     model: str,
     num_queries: int,
     counter_callback: Optional[Callable[[int, int, str], None]] = None,
-    temperature: Optional[float] = LLM_SETTINGS["gpt-4-0125-preview"]["temperature"],
-    max_tokens: Optional[int] = LLM_SETTINGS["gpt-4-0125-preview"]["default_max_tokens"],
+    temperature: Optional[float] = LLM_SETTINGS["claude-3-sonnet-20240229"]["temperature"],
+    max_tokens: Optional[int] = LLM_SETTINGS["claude-3-sonnet-20240229"]["default_max_tokens"],
 ) -> List[str]:
     """Generate multiple queries for fetching information from the web."""
     url_query_prompt = URL_QUERY_PROMPT.format(
@@ -544,8 +552,8 @@ def fetch_additional_information(
     source_links: Optional[List[str]] = None,
     num_urls: Optional[int] = DEFAULT_NUM_URLS,
     num_queries: Optional[int] = DEFAULT_NUM_QUERIES,
-    temperature: Optional[float] = LLM_SETTINGS["gpt-4-0125-preview"]["temperature"],
-    max_tokens: Optional[int] = LLM_SETTINGS["gpt-4-0125-preview"]["default_max_tokens"],
+    temperature: Optional[float] = LLM_SETTINGS["claude-3-sonnet-20240229"]["temperature"],
+    max_tokens: Optional[int] = LLM_SETTINGS["claude-3-sonnet-20240229"]["default_max_tokens"],
 ) -> Tuple[str, Callable[[int, int, str], None]]:
     """Fetch additional information to help answer the user prompt."""
 
@@ -668,13 +676,15 @@ def parser_prediction_response(response: str) -> str:
 
 def run(**kwargs) -> Tuple[Optional[str], Any, Optional[Dict[str, Any]], Any]:
     """Run the task"""
+    tool = kwargs["tool"]
+    model = kwargs.get("model")
+    if "claude" in tool: # maintain backwards compatibility
+        model = "claude-3-sonnet-20240229" 
+    print(f"MODEL: {model}")
     with LLMClientManager(
-        kwargs["api_keys"], kwargs["llm_provider"], embedding_provider="openai"
+        kwargs["api_keys"], model, embedding_provider="openai"
     ):
-        tool = kwargs["tool"]
         prompt = extract_question(kwargs["prompt"])
-        model = kwargs.get("model")
-        print(f"MODEL: {model}")
         max_tokens = kwargs.get(
             "max_tokens", LLM_SETTINGS[model]["default_max_tokens"]
         )

packages/napthaai/customs/prediction_request_rag_cohere/component.yaml

@@ -8,7 +8,7 @@ license: Apache-2.0
 aea_version: '>=1.0.0, <2.0.0'
 fingerprint:
   __init__.py: bafybeiekjzoy2haayvkiwhb2u2epflpqxticud34mma3gdhfzgu36lxwiq
-  prediction_request_rag_cohere.py: bafybeig4oq3tdjuz2la2pz232u5m7347q7gplu5pw4vebbxteuiqw6hh3u
+  prediction_request_rag_cohere.py: bafybeib4jviue2jqktqbxca4gtzxrvvxi5oihhsbvarymiqyp3xkee7soi
 fingerprint_ignore_patterns: []
 entry_point: prediction_request_rag_cohere.py
 callable: run

packages/napthaai/customs/prediction_request_rag_cohere/prediction_request_rag_cohere.py

@@ -40,11 +40,16 @@ class LLMClientManager:
     """Client context manager for LLMs."""
 
     def __init__(
-        self, api_keys: List, llm_provider: str = None, embedding_provider: str = None
+        self, api_keys: List, model: str = None, embedding_provider: str = None
     ):
         self.api_keys = api_keys
-        self.llm_provider = llm_provider
         self.embedding_provider = embedding_provider
+        if "gpt" in model:
+            self.llm_provider = "openai"
+        elif "claude" in model:
+            self.llm_provider = "anthropic"
+        else:
+            self.llm_provider = "openrouter"
 
     def __enter__(self):
         clients = []
@@ -642,13 +647,13 @@ def parser_prediction_response(response: str) -> str:
 
 def run(**kwargs) -> Tuple[Optional[str], Any, Optional[Dict[str, Any]], Any]:
     """Run the task"""
+    model = kwargs.get("model")
+    print(f"MODEL: {model}")
     with LLMClientManager(
-        kwargs["api_keys"], kwargs["llm_provider"], embedding_provider="openai"
+        kwargs["api_keys"], model, embedding_provider="openai"
     ):
         tool = kwargs["tool"]
-        model = kwargs.get("model")
         prompt = extract_question(kwargs["prompt"])
-        print(f"MODEL: {model}")
         max_tokens = kwargs.get(
             "max_tokens", LLM_SETTINGS[model]["default_max_tokens"]
         )

packages/napthaai/customs/prediction_request_reasoning/component.yaml

@@ -7,12 +7,12 @@ license: Apache-2.0
 aea_version: '>=1.0.0, <2.0.0'
 fingerprint:
   __init__.py: bafybeib36ew6vbztldut5xayk5553rylrq7yv4cpqyhwc5ktvd4cx67vwu
-  prediction_request_reasoning.py: bafybeidiabgnlc453spgrdn7rhhl2xc3aa6zqeukkw2bthndbugtjf6bya
+  prediction_request_reasoning.py: bafybeidb43nygtvbhimnsd223ddpoii46dwirb5znmp2g473u4jii36jqa
 fingerprint_ignore_patterns: []
 entry_point: prediction_request_reasoning.py
 callable: run
 params:
-  default_model: claude-3-sonnet-20240229
+  default_model: gpt-4-0125-preview
 dependencies:
   google-api-python-client:
     version: ==2.95.0

packages/napthaai/customs/prediction_request_reasoning/prediction_request_reasoning.py

@@ -43,11 +43,16 @@ class LLMClientManager:
     """Client context manager for LLMs."""
 
     def __init__(
-        self, api_keys: List, llm_provider: str = None, embedding_provider: str = None
+        self, api_keys: List, model: str = None, embedding_provider: str = None
     ):
         self.api_keys = api_keys
-        self.llm_provider = llm_provider
         self.embedding_provider = embedding_provider
+        if "gpt" in model:
+            self.llm_provider = "openai"
+        elif "claude" in model:
+            self.llm_provider = "anthropic"
+        else:
+            self.llm_provider = "openrouter"
 
     def __enter__(self):
         clients = []
@@ -223,6 +228,9 @@ def embeddings(self, model, input):
 }
 ALLOWED_TOOLS = [
     "prediction-request-reasoning",
+
+    # LEGACY
+    "prediction-request-reasoning-claude",
 ]
 ALLOWED_MODELS = list(LLM_SETTINGS.keys())
 DEFAULT_NUM_URLS = defaultdict(lambda: 3)
@@ -834,13 +842,15 @@ def extract_question(prompt: str) -> str:
 
 def run(**kwargs) -> Tuple[str, Optional[str], Optional[Dict[str, Any]], Any]:
     """Run the task"""
+    tool = kwargs["tool"]
+    model = kwargs.get("model")
+    if "claude" in tool: # maintain backwards compatibility
+        model = "claude-3-sonnet-20240229" 
+    print(f"MODEL: {model}")
     with LLMClientManager(
-        kwargs["api_keys"], kwargs["llm_provider"], embedding_provider="openai"
+        kwargs["api_keys"], model, embedding_provider="openai"
     ):
-        tool = kwargs["tool"]
-        model = kwargs.get("model")
         prompt = extract_question(kwargs["prompt"])
-        print(f"MODEL: {model}")
         max_tokens = kwargs.get(
             "max_tokens", LLM_SETTINGS[model]["default_max_tokens"]
         )

packages/napthaai/customs/prediction_url_cot/component.yaml

@@ -7,7 +7,7 @@ license: Apache-2.0
 aea_version: '>=1.0.0, <2.0.0'
 fingerprint:
   __init__.py: bafybeiflni5dkn5fqe7fnu4lgbqxzfrgochhqfbgzwz3vlf5grijp3nkpm
-  prediction_url_cot.py: bafybeifrkisrrphzyhqnvjqtwxynue7xlsmlqhggm5vcuceod2sl4td7ei
+  prediction_url_cot.py: bafybeihebxfv4xj22nq4mkch6xuddcnu7jv473zec2n5p65oxy63asjudy
 fingerprint_ignore_patterns: []
 entry_point: prediction_url_cot.py
 callable: run

packages/napthaai/customs/prediction_url_cot/prediction_url_cot.py

@@ -19,11 +19,16 @@ class LLMClientManager:
     """Client context manager for LLMs."""
 
     def __init__(
-        self, api_keys: List, llm_provider: str = None, embedding_provider: str = None
+        self, api_keys: List, model: str = None, embedding_provider: str = None
     ):
         self.api_keys = api_keys
-        self.llm_provider = llm_provider
         self.embedding_provider = embedding_provider
+        if "gpt" in model:
+            self.llm_provider = "openai"
+        elif "claude" in model:
+            self.llm_provider = "anthropic"
+        else:
+            self.llm_provider = "openrouter"
 
     def __enter__(self):
         clients = []
@@ -180,6 +185,9 @@ def embeddings(self, model, input):
 }
 ALLOWED_TOOLS = [
     "prediction-url-cot",
+
+    # LEGACY
+    "prediction-url-cot-claude",
 ]
 ALLOWED_MODELS = list(LLM_SETTINGS.keys())
 NUM_QUERIES = 5
@@ -583,11 +591,13 @@ def parser_prediction_response(response: str) -> str:
 
 def run(**kwargs) -> Tuple[Optional[str], Any, Optional[Dict[str, Any]], Any]:
     """Run the task"""
-    with LLMClientManager(kwargs["api_keys"], kwargs["llm_provider"]):
-        tool = kwargs["tool"]
-        model = kwargs.get("model")
+    tool = kwargs["tool"]
+    model = kwargs.get("model")
+    if "claude" in tool: # maintain backwards compatibility
+        model = "claude-3-sonnet-20240229" 
+    print(f"MODEL: {model}")
+    with LLMClientManager(kwargs["api_keys"], model):
         prompt = extract_question(kwargs["prompt"])
-        print(f"MODEL: {model}")
         max_tokens = kwargs.get(
             "max_tokens", LLM_SETTINGS[model]["default_max_tokens"]
         )

packages/packages.json

@@ -5,23 +5,23 @@
         "custom/valory/openai_request/0.1.0": "bafybeihjtddwwkvwzaltk6yhtkk3xxnwnkurdtyuy6ki5tpf7h5htvuxnq",
         "custom/valory/prediction_request_embedding/0.1.0": "bafybeifnz5fzxvzyj3mmjpfsre3nzbdieuyjvnxqxuplopp5taz4qw7ys4",
         "custom/valory/resolve_market/0.1.0": "bafybeiaag2e7rsdr3bwg6mlmfyom4vctsdapohco7z45pxhzjymepz3rya",
-        "custom/valory/prediction_request/0.1.0": "bafybeieq5lu3dtgz7svxr6eelbopyvravg3iiomvvtdv33ej5w7hgbjhja",
+        "custom/valory/prediction_request/0.1.0": "bafybeibnshbgciu6inzdjzxeysrwvsin4iitkgd4fkj7a2omjzbdrga2ue",
         "custom/valory/stability_ai_request/0.1.0": "bafybeicyyteycvzj4lk33p4t7mspfarc5d5ktbysu7oqkv6woo4aouxira",
         "custom/polywrap/prediction_with_research_report/0.1.0": "bafybeiewbcbfyjnyqyp4oou6ianxseakblwjyck22bd2doqojjk37uyxwy",
         "custom/jhehemann/prediction_sum_url_content/0.1.0": "bafybeiby55g53cvc4vpbgww5awrlf6x67h7q7pg5xlhwber75ejdkh4twa",
         "custom/psouranis/optimization_by_prompting/0.1.0": "bafybeihb3pyk5qcbj5ib7377p65tznzdsnwilyyhlkcvaj2scmfcpsh6ru",
         "custom/nickcom007/sme_generation_request/0.1.0": "bafybeibqv4ru4lpufy2hvcb3swqhzuq2kejjxmlyepofx6l6mxce6lhiqq",
         "custom/nickcom007/prediction_request_sme/0.1.0": "bafybeigsszaat6k5m5a3ljyem7xdhjflpcm24imtcscgst3tghpwhamglu",
         "custom/napthaai/resolve_market_reasoning/0.1.0": "bafybeiewdqtfkee3od5kuktrhyzexy7466ea3w3to7vv6qnli6qutfrqaa",
-        "custom/napthaai/prediction_request_rag/0.1.0": "bafybeigb7hfqcuvkvsc54526hxwhl6utfj44dnbwiyabcdbghlr5ctkwuu",
-        "custom/napthaai/prediction_request_reasoning/0.1.0": "bafybeiati546f5fyhtwv6yo7zaq3xwtb635p3jp3h3f546stknpbkkyhou",
+        "custom/napthaai/prediction_request_rag/0.1.0": "bafybeif7ufhrlhpuegm6kpiw6jzye6jmp4fjvxgn3hwcv4vkolrrrmidmy",
+        "custom/napthaai/prediction_request_reasoning/0.1.0": "bafybeifzkvc6j5wbbremt2jqig4ozaackzpz3o5okkoihmm3wdpptpviz4",
         "custom/valory/prepare_tx/0.1.0": "bafybeibjqckeb73df724lr4xkrmeh3woqwas4mswa7au65xnwag2edad2e",
         "custom/valory/short_maker/0.1.0": "bafybeif63rt4lkopu3rc3l7sg6tebrrwg2lxqufjx6dx4hoda5yzax43fa",
-        "custom/napthaai/prediction_url_cot/0.1.0": "bafybeidk6s4nqtow6dxmslhjtxzbbnhzpeogyy33e2zpjmqdjijtqb6rz4",
+        "custom/napthaai/prediction_url_cot/0.1.0": "bafybeic3ch7wfhxqvwgoud7xotuu3khs4xch3ej35kox2gulya2hv65wbu",
         "custom/napthaai/prediction_url_cot_claude/0.1.0": "bafybeicbjywni5hx5ssoiv6tnnjbqzsck6cmtsdpr6m562z6afogz5eh44",
         "custom/napthaai/prediction_request_reasoning_claude/0.1.0": "bafybeihtx2cejxoy42jwk2i5m4evfzz537aic5njuawxnzdzwlo63kdduq",
         "custom/napthaai/prediction_request_rag_claude/0.1.0": "bafybeickr32t7nmapuoymjyo3cf5rr2v2zapksxcivuqsgjr2gn6zo6y7y",
-        "custom/napthaai/prediction_request_rag_cohere/0.1.0": "bafybeig3xsmmb4bgbjong6uzvnurf4mwdisqwp3eidmeuo7hj42wkcbymm",
+        "custom/napthaai/prediction_request_rag_cohere/0.1.0": "bafybeie2qi27usujclcje524qu4w6iv3viouq3pxhs2yft3aw26nnerla4",
         "protocol/valory/acn_data_share/0.1.0": "bafybeih5ydonnvrwvy2ygfqgfabkr47s4yw3uqxztmwyfprulwfsoe7ipq",
         "protocol/valory/websocket_client/0.1.0": "bafybeifjk254sy65rna2k32kynzenutujwqndap2r222afvr3zezi27mx4",
         "contract/valory/agent_mech/0.1.0": "bafybeiah6b5epo2hlvzg5rr2cydgpp2waausoyrpnoarf7oa7bw33rex34",

packages/valory/customs/prediction_request/component.yaml

@@ -7,7 +7,7 @@ license: Apache-2.0
 aea_version: '>=1.0.0, <2.0.0'
 fingerprint:
   __init__.py: bafybeibbn67pnrrm4qm3n3kbelvbs3v7fjlrjniywmw2vbizarippidtvi
-  prediction_request.py: bafybeif3s6wd3gotqpg6qdcs7zjszhffkyffeb554r2j5xvtmrbsxy7oca
+  prediction_request.py: bafybeigf5k62mxbmcrvjvsnixpbn3hvxlp2l62sk7jtx5vs7fdg5cgtfxe
 fingerprint_ignore_patterns: []
 entry_point: prediction_request.py
 callable: run

packages/valory/customs/prediction_request/prediction_request.py

@@ -44,9 +44,14 @@
 class LLMClientManager:
     """Client context manager for LLMs."""
 
-    def __init__(self, api_keys: List, llm_provider: str = None):
+    def __init__(self, api_keys: List, model: str = None):
         self.api_keys = api_keys
-        self.llm_provider = llm_provider
+        if "gpt" in model:
+            self.llm_provider = "openai"
+        elif "claude" in model:
+            self.llm_provider = "anthropic"
+        else:
+            self.llm_provider = "openrouter"
 
     def __enter__(self):
         global client
@@ -219,6 +224,10 @@ def count_tokens(text: str, model: str) -> int:
     "prediction-offline",
     "prediction-online",
     # "prediction-online-summarized-info",
+
+    # LEGACY
+    "claude-prediction-offline",
+    "claude-prediction-online",
 ]
 ALLOWED_MODELS = list(LLM_SETTINGS.keys())
 # the default number of URLs to fetch online information for
@@ -653,11 +662,13 @@ def adjust_additional_information(
 
 def run(**kwargs) -> Tuple[str, Optional[str], Optional[Dict[str, Any]], Any]:
     """Run the task"""
-    with LLMClientManager(kwargs["api_keys"], kwargs["llm_provider"]):
-        tool = kwargs["tool"]
+    tool = kwargs["tool"]
+    engine = kwargs.get("model")
+    if "claude" in tool: # maintain backwards compatibility
+        engine = "claude-3-sonnet-20240229" 
+    print(f"ENGINE: {engine}")
+    with LLMClientManager(kwargs["api_keys"], engine):
         prompt = kwargs["prompt"]
-        engine = kwargs.get("model")
-        print(f"ENGINE: {engine}")
         max_tokens = kwargs.get(
             "max_tokens", LLM_SETTINGS[engine]["default_max_tokens"]
         )