Finetuned OpenAI models cost calculation #11715 (#12190)

**Description:**
Add cost calculation for fine tuned models (new and legacy), this is
required after OpenAI added new models for fine tuning and separated the
costs of I/O for fine tuned models.
Also I updated the relevant unit tests
see https://platform.openai.com/docs/guides/fine-tuning for more
information.
issue: https://github.com/langchain-ai/langchain/issues/11715

  - **Issue:** 11715
  - **Twitter handle:** @nirkopler

libs/langchain/langchain/callbacks/openai_info.py

@@ -57,10 +57,19 @@ MODEL_COST_PER_1K_TOKENS = {
     "text-davinci-003": 0.02,
     "text-davinci-002": 0.02,
     "code-davinci-002": 0.02,
-    "ada-finetuned": 0.0016,
-    "babbage-finetuned": 0.0024,
-    "curie-finetuned": 0.012,
-    "davinci-finetuned": 0.12,
+    # Fine Tuned input
+    "babbage-002-finetuned": 0.0016,
+    "davinci-002-finetuned": 0.012,
+    "gpt-3.5-turbo-0613-finetuned": 0.012,
+    # Fine Tuned output
+    "babbage-002-finetuned-completion": 0.0016,
+    "davinci-002-finetuned-completion": 0.012,
+    "gpt-3.5-turbo-0613-finetuned-completion": 0.016,
+    # Legacy fine-tuned models
+    "ada-finetuned-legacy": 0.0016,
+    "babbage-finetuned-legacy": 0.0024,
+    "curie-finetuned-legacy": 0.012,
+    "davinci-finetuned-legacy": 0.12,
 }
 
 
@@ -82,11 +91,14 @@ def standardize_model_name(
     """
     model_name = model_name.lower()
     if "ft-" in model_name:
-        return model_name.split(":")[0] + "-finetuned"
+        return model_name.split(":")[0] + "-finetuned-legacy"
+    if "ft:" in model_name:
+        return model_name.split(":")[1] + "-finetuned"
     elif is_completion and (
         model_name.startswith("gpt-4")
         or model_name.startswith("gpt-3.5")
         or model_name.startswith("gpt-35")
+        or ("finetuned" in model_name and "legacy" not in model_name)
     ):
         return model_name + "-completion"
     else:

libs/langchain/tests/unit_tests/callbacks/test_openai_info.py

@@ -49,7 +49,21 @@ def test_on_llm_end_custom_model(handler: OpenAICallbackHandler) -> None:
     assert handler.total_cost == 0
 
 
-def test_on_llm_end_finetuned_model(handler: OpenAICallbackHandler) -> None:
+@pytest.mark.parametrize(
+    "model_name",
+    [
+        "ada:ft-your-org:custom-model-name-2022-02-15-04-21-04",
+        "babbage:ft-your-org:custom-model-name-2022-02-15-04-21-04",
+        "curie:ft-your-org:custom-model-name-2022-02-15-04-21-04",
+        "davinci:ft-your-org:custom-model-name-2022-02-15-04-21-04",
+        "ft:babbage-002:your-org:custom-model-name:1abcdefg",
+        "ft:davinci-002:your-org:custom-model-name:1abcdefg",
+        "ft:gpt-3.5-turbo-0613:your-org:custom-model-name:1abcdefg",
+    ],
+)
+def test_on_llm_end_finetuned_model(
+    handler: OpenAICallbackHandler, model_name: str
+) -> None:
     response = LLMResult(
         generations=[],
         llm_output={
@@ -58,7 +72,7 @@ def test_on_llm_end_finetuned_model(handler: OpenAICallbackHandler) -> None:
                 "completion_tokens": 1,
                 "total_tokens": 3,
             },
-            "model_name": "ada:ft-your-org:custom-model-name-2022-02-15-04-21-04",
+            "model_name": model_name,
         },
     )
     handler.on_llm_end(response)