payi 0.1.0a109__tar.gz → 0.1.0a111__tar.gz

payi-0.1.0a111/.release-please-manifest.json

@@ -0,0 +1,3 @@
+{
+  ".": "0.1.0-alpha.111"
+}

{payi-0.1.0a109 → payi-0.1.0a111}/CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## 0.1.0-alpha.111 (2025-08-18)
+
+Full Changelog: [v0.1.0-alpha.110...v0.1.0-alpha.111](https://github.com/Pay-i/pay-i-python/compare/v0.1.0-alpha.110...v0.1.0-alpha.111)
+
+### Features
+
+* anthropic large context, cohere english v3 embed, amazon titan v1 text embed ([#356](https://github.com/Pay-i/pay-i-python/issues/356)) ([9a3549e](https://github.com/Pay-i/pay-i-python/commit/9a3549e1abd284d1647d7c63fa3cdff01ac14a98))
+
+## 0.1.0-alpha.110 (2025-08-09)
+
+Full Changelog: [v0.1.0-alpha.109...v0.1.0-alpha.110](https://github.com/Pay-i/pay-i-python/compare/v0.1.0-alpha.109...v0.1.0-alpha.110)
+
+### Chores
+
+* **internal:** update comment in script ([62f3d00](https://github.com/Pay-i/pay-i-python/commit/62f3d00cbe71bc217f13b9e354a53b9d42dd8c27))
+* update @stainless-api/prism-cli to v5.15.0 ([87fc0ef](https://github.com/Pay-i/pay-i-python/commit/87fc0ef32878b5ba1c4003d4c392ce31f8cfde86))
+
 ## 0.1.0-alpha.109 (2025-08-06)
 
 Full Changelog: [v0.1.0-alpha.108...v0.1.0-alpha.109](https://github.com/Pay-i/pay-i-python/compare/v0.1.0-alpha.108...v0.1.0-alpha.109)

{payi-0.1.0a109 → payi-0.1.0a111}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: payi
-Version: 0.1.0a109
+Version: 0.1.0a111
 Summary: The official Python library for the payi API
 Project-URL: Homepage, https://github.com/Pay-i/pay-i-python
 Project-URL: Repository, https://github.com/Pay-i/pay-i-python
@@ -29,6 +29,7 @@ Requires-Dist: nest-asyncio>=1.6.0
 Requires-Dist: pydantic<3,>=1.9.0
 Requires-Dist: sniffio
 Requires-Dist: tiktoken>=0.8.0
+Requires-Dist: tokenizers>=0.21.4
 Requires-Dist: typing-extensions<5,>=4.10
 Requires-Dist: wrapt>=1.17.2
 Provides-Extra: aiohttp

{payi-0.1.0a109 → payi-0.1.0a111}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "payi"
-version = "0.1.0-alpha.109"
+version = "0.1.0-alpha.111"
 description = "The official Python library for the payi API"
 dynamic = ["readme"]
 license = "Apache-2.0"
@@ -17,6 +17,7 @@ dependencies = [
     "wrapt>=1.17.2",
     "tiktoken>=0.8.0",
     "nest-asyncio>=1.6.0",
+    "tokenizers>=0.21.4"
 ]
 requires-python = ">= 3.8"
 classifiers = [

{payi-0.1.0a109 → payi-0.1.0a111}/requirements-dev.lock

@@ -47,12 +47,17 @@ exceptiongroup==1.2.2
 execnet==2.1.1
     # via pytest-xdist
 filelock==3.12.4
+    # via huggingface-hub
     # via virtualenv
 frozenlist==1.6.2
     # via aiohttp
     # via aiosignal
+fsspec==2025.7.0
+    # via huggingface-hub
 h11==0.16.0
     # via httpcore
+hf-xet==1.1.7
+    # via huggingface-hub
 httpcore==1.0.9
     # via httpx
 httpx==0.28.1
@@ -61,6 +66,8 @@ httpx==0.28.1
     # via respx
 httpx-aiohttp==0.1.8
     # via payi
+huggingface-hub==0.34.4
+    # via tokenizers
 idna==3.4
     # via anyio
     # via httpx
@@ -85,6 +92,7 @@ nodeenv==1.8.0
     # via pyright
 nox==2023.4.22
 packaging==23.2
+    # via huggingface-hub
     # via nox
     # via pytest
 platformdirs==3.11.0
@@ -110,9 +118,12 @@ python-dateutil==2.8.2
     # via time-machine
 pytz==2023.3.post1
     # via dirty-equals
+pyyaml==6.0.2
+    # via huggingface-hub
 regex==2024.11.6
     # via tiktoken
 requests==2.32.3
+    # via huggingface-hub
     # via tiktoken
 respx==0.22.0
 rich==13.7.1
@@ -127,11 +138,16 @@ sniffio==1.3.0
 tiktoken==0.8.0
     # via payi
 time-machine==2.9.0
+tokenizers==0.21.4
+    # via payi
 tomli==2.0.2
     # via mypy
     # via pytest
+tqdm==4.67.1
+    # via huggingface-hub
 typing-extensions==4.12.2
     # via anyio
+    # via huggingface-hub
     # via multidict
     # via mypy
     # via payi

{payi-0.1.0a109 → payi-0.1.0a111}/requirements.lock

@@ -36,11 +36,17 @@ distro==1.8.0
     # via payi
 exceptiongroup==1.2.2
     # via anyio
+filelock==3.19.1
+    # via huggingface-hub
 frozenlist==1.6.2
     # via aiohttp
     # via aiosignal
+fsspec==2025.7.0
+    # via huggingface-hub
 h11==0.16.0
     # via httpcore
+hf-xet==1.1.7
+    # via huggingface-hub
 httpcore==1.0.9
     # via httpx
 httpx==0.28.1
@@ -48,6 +54,8 @@ httpx==0.28.1
     # via payi
 httpx-aiohttp==0.1.8
     # via payi
+huggingface-hub==0.34.4
+    # via tokenizers
 idna==3.4
     # via anyio
     # via httpx
@@ -58,6 +66,8 @@ multidict==6.4.4
     # via yarl
 nest-asyncio==1.6.0
     # via payi
+packaging==25.0
+    # via huggingface-hub
 propcache==0.3.1
     # via aiohttp
     # via yarl
@@ -65,17 +75,25 @@ pydantic==2.10.3
     # via payi
 pydantic-core==2.27.1
     # via pydantic
+pyyaml==6.0.2
+    # via huggingface-hub
 regex==2024.11.6
     # via tiktoken
 requests==2.32.3
+    # via huggingface-hub
     # via tiktoken
 sniffio==1.3.0
     # via anyio
     # via payi
 tiktoken==0.8.0
     # via payi
+tokenizers==0.21.4
+    # via payi
+tqdm==4.67.1
+    # via huggingface-hub
 typing-extensions==4.12.2
     # via anyio
+    # via huggingface-hub
     # via multidict
     # via payi
     # via pydantic

{payi-0.1.0a109 → payi-0.1.0a111}/src/payi/_version.py

@@ -1,4 +1,4 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
 __title__ = "payi"
-__version__ = "0.1.0-alpha.109"  # x-release-please-version
+__version__ = "0.1.0-alpha.111"  # x-release-please-version

{payi-0.1.0a109 → payi-0.1.0a111}/src/payi/lib/AnthropicInstrumentor.py

@@ -45,6 +45,18 @@ class AnthropicInstrumentor:
                 stream_messages_wrapper(instrumentor),
             )
 
+            wrap_function_wrapper(
+                "anthropic.resources.beta.messages",
+                "Messages.create",
+                messages_wrapper(instrumentor),
+            )
+
+            wrap_function_wrapper(
+                "anthropic.resources.beta.messages",
+                "Messages.stream",
+                stream_messages_wrapper(instrumentor),
+            )
+
             wrap_function_wrapper(
                 "anthropic.resources.messages",
                 "AsyncMessages.create",
@@ -57,6 +69,18 @@ class AnthropicInstrumentor:
                 astream_messages_wrapper(instrumentor),
             )
 
+            wrap_function_wrapper(
+                "anthropic.resources.beta.messages",
+                "AsyncMessages.create",
+                amessages_wrapper(instrumentor),
+            )
+
+            wrap_function_wrapper(
+                "anthropic.resources.beta.messages",
+                "AsyncMessages.stream",
+                astream_messages_wrapper(instrumentor),
+            )
+
         except Exception as e:
             instrumentor._logger.debug(f"Error instrumenting anthropic: {e}")
             return
@@ -220,23 +244,52 @@ class _AnthropicProviderRequest(_ProviderRequest):
 
         return True
 
-def anthropic_process_synchronous_response(request: _ProviderRequest, response: 'dict[str, Any]', log_prompt_and_response: bool, assign_id: bool) -> Any:
-    usage = response['usage']
+def anthropic_process_compute_input_cost(request: _ProviderRequest, usage: 'dict[str, Any]') -> int:
     input = usage['input_tokens']
-    output = usage['output_tokens']
     units: dict[str, Units] = request._ingest["units"]
 
     cache_creation_input_tokens = usage.get("cache_creation_input_tokens", 0)
-    if cache_creation_input_tokens > 0:
-        units["text_cache_write"] = Units(input=cache_creation_input_tokens, output=0)
+    cache_read_input_tokens = usage.get("cache_read_input_tokens", 0)
+
+    total_input_tokens = input + cache_creation_input_tokens + cache_read_input_tokens
+
+    request._is_large_context = total_input_tokens > 200000
+    large_context = "_large_context" if request._is_large_context else ""
+
+    cache_creation: dict[str, int] = usage.get("cache_creation", {})
+    ephemeral_5m_input_tokens: Optional[int] = None
+    ephemeral_1h_input_tokens: Optional[int] = None
+    textCacheWriteAdded = False
+
+    if cache_creation:
+        ephemeral_5m_input_tokens = cache_creation.get("ephemeral_5m_input_tokens", 0)
+        if ephemeral_5m_input_tokens > 0:
+            textCacheWriteAdded = True
+            units["text_cache_write"+large_context] = Units(input=ephemeral_5m_input_tokens, output=0)
+
+        ephemeral_1h_input_tokens = cache_creation.get("ephemeral_1h_input_tokens", 0)
+        if ephemeral_1h_input_tokens > 0:
+            textCacheWriteAdded = True
+            units["text_cache_write_1h"+large_context] = Units(input=ephemeral_1h_input_tokens, output=0)
+
+    if textCacheWriteAdded is False and cache_creation_input_tokens > 0:
+        units["text_cache_write"+large_context] = Units(input=cache_creation_input_tokens, output=0)
 
     cache_read_input_tokens = usage.get("cache_read_input_tokens", 0)
     if cache_read_input_tokens > 0:
-        units["text_cache_read"] = Units(input=cache_read_input_tokens, output=0)
+        units["text_cache_read"+large_context] = Units(input=cache_read_input_tokens, output=0)
 
-    input = _PayiInstrumentor.update_for_vision(input, units, request._estimated_prompt_tokens)
+    return _PayiInstrumentor.update_for_vision(input, units, request._estimated_prompt_tokens, is_large_context=request._is_large_context)
 
-    units["text"] = Units(input=input, output=output)
+def anthropic_process_synchronous_response(request: _ProviderRequest, response: 'dict[str, Any]', log_prompt_and_response: bool, assign_id: bool) -> Any:
+    usage = response['usage']
+    units: dict[str, Units] = request._ingest["units"]
+
+    input_tokens = anthropic_process_compute_input_cost(request, usage)
+    output = usage['output_tokens']
+
+    large_context = "_large_context" if request._is_large_context else ""
+    units["text"+large_context] = Units(input=input_tokens, output=output)
 
     content = response.get('content', [])
     if content:
@@ -277,31 +330,25 @@ def anthropic_process_chunk(request: _ProviderRequest, chunk: 'dict[str, Any]',
         usage = message['usage']
         units = request._ingest["units"]
 
-        input = _PayiInstrumentor.update_for_vision(usage['input_tokens'], units, request._estimated_prompt_tokens)
-
-        units["text"] = Units(input=input, output=0)
-
-        text_cache_write: int = usage.get("cache_creation_input_tokens", 0)
-        if text_cache_write > 0:
-            units["text_cache_write"] = Units(input=text_cache_write, output=0)
+        input = anthropic_process_compute_input_cost(request, usage)
 
-        text_cache_read: int = usage.get("cache_read_input_tokens", 0)
-        if text_cache_read > 0:
-            units["text_cache_read"] = Units(input=text_cache_read, output=0)
+        large_context = "_large_context" if request._is_large_context else ""
+        units["text"+large_context] = Units(input=input, output=0)
 
         request._instrumentor._logger.debug(f"Anthropic streaming captured {input} input tokens, ")
 
     elif type == "message_delta":
         usage = chunk.get('usage', {})
         ingest = True
+        large_context = "_large_context" if request._is_large_context else ""
 
         # Web search will return an updated input tokens value at the end of streaming
         input_tokens = usage.get('input_tokens', None)
         if input_tokens is not None:
             request._instrumentor._logger.debug(f"Anthropic streaming finished, updated input tokens: {input_tokens}")
-            request._ingest["units"]["text"]["input"] = input_tokens
+            request._ingest["units"]["text"+large_context]["input"] = input_tokens
 
-        request._ingest["units"]["text"]["output"] = usage.get('output_tokens', 0)
+        request._ingest["units"]["text"+large_context]["output"] = usage.get('output_tokens', 0)
 
         request._instrumentor._logger.debug(f"Anthropic streaming finished: output tokens {usage.get('output_tokens', 0)} ")
 

{payi-0.1.0a109 → payi-0.1.0a111}/src/payi/lib/BedrockInstrumentor.py

@@ -5,6 +5,7 @@ from functools import wraps
 from typing_extensions import override
 
 from wrapt import ObjectProxy, wrap_function_wrapper  # type: ignore
+from tokenizers import Tokenizer  # type: ignore
 
 from payi.lib.helpers import PayiCategories, PayiHeaderNames, payi_aws_bedrock_url
 from payi.types.ingest_units_params import Units
@@ -102,6 +103,8 @@ def _redirect_to_payi(request: Any, event_name: str, **_: 'dict[str, Any]') -> N
 
 
 class InvokeResponseWrapper(ObjectProxy): # type: ignore
+    _cohere_embed_english_v3_tokenizer: Optional[Tokenizer] = None
+
     def __init__(
         self,
         response: Any,
@@ -160,6 +163,25 @@ class InvokeResponseWrapper(ObjectProxy): # type: ignore
 
             bedrock_converse_process_synchronous_function_call(self._request, response)
 
+        elif self._request._is_amazon_titan_embed_text_v1:
+            input = response.get('inputTextTokenCount', 0)
+            units["text"] = Units(input=input, output=0)
+
+        elif self._request._is_cohere_embed_english_v3:
+            texts: list[str] = response.get("texts", [])
+            if texts and len(texts) > 0:
+                text = " ".join(texts)
+
+                if self._cohere_embed_english_v3_tokenizer is None:
+                    current_dir = os.path.dirname(os.path.abspath(__file__))
+                    tokenizer_path = os.path.join(current_dir, "data", "cohere_embed_english_v3.json")
+                    self._cohere_embed_english_v3_tokenizer = Tokenizer.from_file(tokenizer_path) # type: ignore
+
+                tokens: list = self._cohere_embed_english_v3_tokenizer.encode(text, add_special_tokens=False).tokens # type: ignore
+
+                if tokens and isinstance(tokens, list):
+                    units["text"] = Units(input=len(tokens), output=0) # type: ignore
+
         if self._log_prompt_and_response:
             ingest["provider_response_json"] = data.decode('utf-8') # type: ignore
             
@@ -287,6 +309,8 @@ class _BedrockInvokeProviderRequest(_BedrockProviderRequest):
         self._is_anthropic: bool = 'anthropic' in model_id
         self._is_nova: bool = 'nova' in model_id
         self._is_meta: bool = 'meta' in model_id
+        self._is_amazon_titan_embed_text_v1: bool = 'amazon.titan-embed-text-v1' == model_id
+        self._is_cohere_embed_english_v3: bool = 'cohere.embed-english-v3' == model_id
 
     @override
     def process_request(self, instance: Any, extra_headers: 'dict[str, str]', args: Sequence[Any], kwargs: Any) -> bool:
@@ -302,7 +326,17 @@ class _BedrockInvokeProviderRequest(_BedrockProviderRequest):
                     anthropic_has_image_and_get_texts(self, messages)
             except Exception as e:
                 self._instrumentor._logger.debug(f"Bedrock invoke error processing request body: {e}")
-
+        elif self._is_cohere_embed_english_v3:
+            try:
+                body = json.loads( kwargs.get("body", ""))
+                input_type = body.get("input_type", "")
+                if input_type == 'image':
+                    images = body.get("images", [])
+                    if (len(images) > 0):
+                        # only supports one image according to docs
+                        self._ingest["units"]["vision"] = Units(input=1, output=0)
+            except Exception as e:
+                self._instrumentor._logger.debug(f"Bedrock invoke error processing request body: {e}")
         return True
 
     @override

{payi-0.1.0a109 → payi-0.1.0a111}/src/payi/lib/VertexRequest.py

@@ -148,7 +148,7 @@ class _VertexRequest(_ProviderRequest): # type: ignore
             return model.startswith("gemini-1.")
 
         def is_large_context_token_model(model: str, input_tokens: int) -> bool:
-            return model.startswith("gemini-2.5-pro") and input_tokens > 200_000
+            return model.startswith("gemini-2.5-pro") and input_tokens > 200000
 
         def add_units(request: _ProviderRequest, key: str, input: Optional[int] = None, output: Optional[int] = None) -> None:
             if key not in request._ingest["units"]:
@@ -172,6 +172,7 @@ class _VertexRequest(_ProviderRequest): # type: ignore
 
         if is_character_billing_model(model):
             if input > 128000: 
+                self._is_large_context = True
                 large_context = "_large_context"
 
             # gemini 1.0 and 1.5 units are reported in characters, per second, per image, etc...
@@ -222,6 +223,7 @@ class _VertexRequest(_ProviderRequest): # type: ignore
             thinking_token_count = usage.get("thoughts_token_count", 0)
 
             if is_large_context_token_model(model, input):
+                self._is_large_context = True
                 large_context = "_large_context"
 
             cache_details: dict[str, int] = {}