traceAI-transformers 0.1.0__tar.gz

traceai_transformers-0.1.0/.gitignore

@@ -0,0 +1,87 @@
+*.py[cod]
+
+# C extensions
+*.so
+
+# IDEs
+.idea/
+.vscode/
+
+# Packages
+*.egg
+*.egg-info
+.eggs/
+# build
+parts
+bin
+var
+sdist
+dist
+develop-eggs
+.installed.cfg
+lib
+lib64
+
+# Installer logs
+pip-log.txt
+
+# Unit test / coverage reports
+.coverage
+.tox
+nosetests.xml
+
+# Complexity
+output/*.html
+output/*/index.html
+
+# Sphinx
+docs/_build
+
+# Cookiecutter
+output/
+
+.workon
+
+**.wav
+**.joblib
+
+**/__pycache__
+
+#Excel files
+*.xlsx
+#csv files
+*.csv
+#virtual environments
+**/.venv
+**/env
+**/.env
+*.DS_Store
+*.pdf
+**/node_modules
+**/dist
+*.log
+.DS_Store
+**/package-lock.json
+# Standalone npm example (outside the pnpm workspace): its lockfile pins the tested dependency set.
+!typescript/examples/tanstack-ai/package-lock.json
+**/tsconfig.tsbuildinfo
+**/.fi
+**/.pytest_cache
+planning/
+
+# Java build artifacts
+**/target/
+**/.settings/
+**/.project
+**/.classpath
+**/.factorypath
+*.class
+*.jar
+*.war
+
+# C# build artifacts
+**/bin/
+**/obj/
+
+# Python
+**/uv.lock

traceai_transformers-0.1.0/PKG-INFO

@@ -0,0 +1,103 @@
+Metadata-Version: 2.5
+Name: traceAI-transformers
+Version: 0.1.0
+Summary: OpenTelemetry instrumentation for local Hugging Face Transformers pipelines
+Project-URL: Repository, https://github.com/future-agi/traceAI/tree/dev/python/frameworks/transformers
+Project-URL: Documentation, https://docs.futureagi.com/docs/tracing
+Author-email: Future AGI <no-reply@futureagi.com>
+License-Expression: Apache-2.0
+Requires-Python: <3.14,>=3.10
+Requires-Dist: fi-instrumentation-otel<2,>=1.2.2
+Requires-Dist: opentelemetry-instrumentation<1,>=0.50b0
+Requires-Dist: transformers<6,>=4.42
+Requires-Dist: wrapt<2,>=1.15
+Provides-Extra: dev
+Requires-Dist: mypy<2,>=1.15; extra == 'dev'
+Requires-Dist: pytest<9,>=8; extra == 'dev'
+Requires-Dist: torch<3,>=2.6; extra == 'dev'
+Requires-Dist: traceai-huggingface>=0.1.0; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# traceAI Transformers
+
+Trace local Hugging Face Transformers pipeline inference in Future AGI. For remote Hugging Face InferenceClient calls, use `traceAI-huggingface` instead; the two instrumentors can run together.
+
+## Install
+
+```bash
+pip install traceAI-transformers torch
+```
+
+Python 3.10–3.13 and Transformers 4.42–5.x are supported. Transformers 5 requires Python 3.11 or later. Use PyTorch pipelines; select the appropriate PyTorch build for your CPU or GPU.
+
+## Trace inference
+
+Set `FI_API_KEY` and `FI_SECRET_KEY` in your environment, then configure tracing before calling your pipeline:
+
+```python
+import os
+from fi_instrumentation import register
+from fi_instrumentation.fi_types import ProjectType
+from traceai_transformers import TransformersInstrumentor
+from transformers import pipeline
+
+provider = register(
+    project_name="local-transformers",
+    project_type=ProjectType.OBSERVE,
+    headers={
+        "FI_API_KEY": os.environ["FI_API_KEY"],
+        "FI_SECRET_KEY": os.environ["FI_SECRET_KEY"],
+    },
+    verbose=False,
+)
+TransformersInstrumentor().instrument(tracer_provider=provider)
+
+generate = pipeline(
+    "text-generation",
+    model="HuggingFaceTB/SmolLM2-135M-Instruct",
+    device=-1,
+)
+result = generate(
+    [{"role": "user", "content": "Explain what a tokenizer does in one sentence."}],
+    max_new_tokens=32,
+    do_sample=False,
+)
+print(result[0]["generated_text"][-1]["content"])
+provider.force_flush()
+provider.shutdown()
+```
+
+The first run downloads the model. `register` exports OTLP traces to `https://api.futureagi.com/tracer/v1/traces` with `FI_API_KEY` and `FI_SECRET_KEY` headers. This is a traces-only endpoint, not a destination for OTLP metrics or logs. Open the `local-transformers` project in Future AGI to inspect `transformers.text-generation` spans.
+
+## Content and privacy
+
+By default, only task, model architecture, timing, status and error information are recorded. Model paths, generation parameters, embedding vectors and token usage estimates are not collected.
+
+To include bounded input/output previews:
+
+```python
+from fi_instrumentation import TraceConfig
+
+TransformersInstrumentor().instrument(
+    tracer_provider=provider,
+    capture_content=True,
+    config=TraceConfig(hide_error_details=True),
+)
+```
+
+Choose this configuration on your first `instrument()` call, or call `uninstrument()` before changing it. Previews are limited to 4,096 characters and may be truncated. `TraceConfig` input/output privacy flags override content capture. Hiding inputs also suppresses output previews, because pipeline responses can contain the original prompt or conversation. Set `FI_HIDE_ERROR_DETAILS=true` to omit exception messages, stack traces and error descriptions. Error details can contain inputs even when content capture is off.
+
+## Supported calls
+
+| Pipeline | Span kind | Versions |
+| --- | --- | --- |
+| Text generation, including chat, batches and streamers | LLM | 4.42–5.x |
+| Feature extraction, including tensor results | EMBEDDING | 4.42–5.x |
+| Text-to-text generation, summarization and translation | LLM | 4.x only |
+| Extractive question answering | CHAIN | 4.x only |
+
+Transformers 5 removed the historical text-to-text, summarization, translation and extractive question-answering pipelines. Use a text-generation chat model on 5.x. Direct `model.generate()` calls, other tasks, TensorFlow pipelines, and model loading are not instrumented.
+
+Generator and dataset calls remain lazy. Each traversal gets a span that starts when the first result is requested and ends on exhaustion, error or explicit `iterator.close()`. Dataset results preserve their length and `DataLoader` compatibility. Starting another traversal preserves the dataset's reset behavior and ends any unfinished span from the previous traversal. The span is current only while producing the next result, not while your application handles it. Close partially consumed iterators to end their spans promptly. Lazy payloads are not collected. Multiprocessing `DataLoader` workers receive the native dataset through serialization, not the instrumentor's tracer provider or parent context; inference in those workers is not traced by this dataset wrapper. To trace inference inside a worker, initialize its own exporter there and add explicit spans around that worker's inference; the parent instrumentor does not trace a deserialized dataset. Streamer spans end when pipeline inference returns, independently of when queued tokens are consumed. When starting a worker thread, propagate OpenTelemetry context explicitly if it should inherit a parent span.
+
+Instrumentation is idempotent and supports `TransformersInstrumentor().uninstrument()` and re-instrumentation. It uses your supplied tracer provider without replacing the global provider.

traceai_transformers-0.1.0/README.md

@@ -0,0 +1,83 @@
+# traceAI Transformers
+
+Trace local Hugging Face Transformers pipeline inference in Future AGI. For remote Hugging Face InferenceClient calls, use `traceAI-huggingface` instead; the two instrumentors can run together.
+
+## Install
+
+```bash
+pip install traceAI-transformers torch
+```
+
+Python 3.10–3.13 and Transformers 4.42–5.x are supported. Transformers 5 requires Python 3.11 or later. Use PyTorch pipelines; select the appropriate PyTorch build for your CPU or GPU.
+
+## Trace inference
+
+Set `FI_API_KEY` and `FI_SECRET_KEY` in your environment, then configure tracing before calling your pipeline:
+
+```python
+import os
+from fi_instrumentation import register
+from fi_instrumentation.fi_types import ProjectType
+from traceai_transformers import TransformersInstrumentor
+from transformers import pipeline
+
+provider = register(
+    project_name="local-transformers",
+    project_type=ProjectType.OBSERVE,
+    headers={
+        "FI_API_KEY": os.environ["FI_API_KEY"],
+        "FI_SECRET_KEY": os.environ["FI_SECRET_KEY"],
+    },
+    verbose=False,
+)
+TransformersInstrumentor().instrument(tracer_provider=provider)
+
+generate = pipeline(
+    "text-generation",
+    model="HuggingFaceTB/SmolLM2-135M-Instruct",
+    device=-1,
+)
+result = generate(
+    [{"role": "user", "content": "Explain what a tokenizer does in one sentence."}],
+    max_new_tokens=32,
+    do_sample=False,
+)
+print(result[0]["generated_text"][-1]["content"])
+provider.force_flush()
+provider.shutdown()
+```
+
+The first run downloads the model. `register` exports OTLP traces to `https://api.futureagi.com/tracer/v1/traces` with `FI_API_KEY` and `FI_SECRET_KEY` headers. This is a traces-only endpoint, not a destination for OTLP metrics or logs. Open the `local-transformers` project in Future AGI to inspect `transformers.text-generation` spans.
+
+## Content and privacy
+
+By default, only task, model architecture, timing, status and error information are recorded. Model paths, generation parameters, embedding vectors and token usage estimates are not collected.
+
+To include bounded input/output previews:
+
+```python
+from fi_instrumentation import TraceConfig
+
+TransformersInstrumentor().instrument(
+    tracer_provider=provider,
+    capture_content=True,
+    config=TraceConfig(hide_error_details=True),
+)
+```
+
+Choose this configuration on your first `instrument()` call, or call `uninstrument()` before changing it. Previews are limited to 4,096 characters and may be truncated. `TraceConfig` input/output privacy flags override content capture. Hiding inputs also suppresses output previews, because pipeline responses can contain the original prompt or conversation. Set `FI_HIDE_ERROR_DETAILS=true` to omit exception messages, stack traces and error descriptions. Error details can contain inputs even when content capture is off.
+
+## Supported calls
+
+| Pipeline | Span kind | Versions |
+| --- | --- | --- |
+| Text generation, including chat, batches and streamers | LLM | 4.42–5.x |
+| Feature extraction, including tensor results | EMBEDDING | 4.42–5.x |
+| Text-to-text generation, summarization and translation | LLM | 4.x only |
+| Extractive question answering | CHAIN | 4.x only |
+
+Transformers 5 removed the historical text-to-text, summarization, translation and extractive question-answering pipelines. Use a text-generation chat model on 5.x. Direct `model.generate()` calls, other tasks, TensorFlow pipelines, and model loading are not instrumented.
+
+Generator and dataset calls remain lazy. Each traversal gets a span that starts when the first result is requested and ends on exhaustion, error or explicit `iterator.close()`. Dataset results preserve their length and `DataLoader` compatibility. Starting another traversal preserves the dataset's reset behavior and ends any unfinished span from the previous traversal. The span is current only while producing the next result, not while your application handles it. Close partially consumed iterators to end their spans promptly. Lazy payloads are not collected. Multiprocessing `DataLoader` workers receive the native dataset through serialization, not the instrumentor's tracer provider or parent context; inference in those workers is not traced by this dataset wrapper. To trace inference inside a worker, initialize its own exporter there and add explicit spans around that worker's inference; the parent instrumentor does not trace a deserialized dataset. Streamer spans end when pipeline inference returns, independently of when queued tokens are consumed. When starting a worker thread, propagate OpenTelemetry context explicitly if it should inherit a parent span.
+
+Instrumentation is idempotent and supports `TransformersInstrumentor().uninstrument()` and re-instrumentation. It uses your supplied tracer provider without replacing the global provider.

traceai_transformers-0.1.0/pyproject.toml

@@ -0,0 +1,34 @@
+[build-system]
+requires = ["hatchling>=1.27,<2"]
+build-backend = "hatchling.build"
+
+[project]
+name = "traceAI-transformers"
+version = "0.1.0"
+description = "OpenTelemetry instrumentation for local Hugging Face Transformers pipelines"
+readme = "README.md"
+license = "Apache-2.0"
+authors = [{name = "Future AGI", email = "no-reply@futureagi.com"}]
+requires-python = ">=3.10,<3.14"
+dependencies = [
+  "fi-instrumentation-otel>=1.2.2,<2",
+  "transformers>=4.42,<6",
+  "opentelemetry-instrumentation>=0.50b0,<1",
+  "wrapt>=1.15,<2",
+]
+
+[project.optional-dependencies]
+dev = ["pytest>=8,<9", "torch>=2.6,<3", "traceAI-huggingface>=0.1.0", "mypy>=1.15,<2"]
+
+[project.urls]
+Repository = "https://github.com/future-agi/traceAI/tree/dev/python/frameworks/transformers"
+Documentation = "https://docs.futureagi.com/docs/tracing"
+
+[project.entry-points.opentelemetry_instrumentor]
+transformers = "traceai_transformers:TransformersInstrumentor"
+
+[tool.hatch.build.targets.wheel]
+packages = ["traceai_transformers"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]