^j-"dZddlmZddlmZmZddlmZddlm Z ddl m Z ddl m Z dd lmZmZd d lmZd d l mZe j(eZe rerdd lZddlmZddlmZGddeZy )z3Abstract base class for all Transformers exporters.) annotations)ABCabstractmethod)MutableMapping) TYPE_CHECKING)version)logging)_is_package_availableis_torch_available)ExportConfigMixin)decompose_for_generationN)Cache)PreTrainedModelc~eZdZUdZgZded<iZded<iZded<dZdZ e d d Z d d Z y ) HfExporterz Abstract base class for all Transformers exporters. Subclass and implement [`~HfExporter.export`] to add a new export backend. z list[str]required_packageszdict[str, str] min_versionstested_versionsc$|jy)N)validate_environment)selfs f/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/transformers/exporters/base.py__init__zHfExporter.__init__7s !!#c gg}}jD]}t|d\}}|s|j|'jj |}|E|dk7sK|j ddd} |j ddd} | | k7s{|j|| | f|r>dj fd |D} td tjd | g} jjD]s\}} t|d\}}|dk(s?tj|j dddtj| ksZ| j|d | d |du| r2ttjddj | |rGdj d|D}tjtjd|dyy)zYCheck `required_packages` are installed and warn on version drift from `tested_versions`.T)return_versionNzN/A+r rz, c3fK|](}|jvr|dj|n|*yw)z==N)r).0pkgrs r z2HfExporter.validate_environment..Ns@be@T@T9T3%r$..s345Z]]s.1zTo use z-, please install the following dependencies: z>=z (found )z requires newer versions of: c38K|]\}}}|d|d|yw)z : installed z , tested N)r!r"gotwants rr#z2HfExporter.validate_environment..]s)e^SRUW[3%|C5 $ Hesu~ is experimental and patches many backend internals; behaviour may differ from what was validated. Version drift detected — z-. If you hit issues, try the tested versions.)rr appendrgetsplitjoin ImportErrortype__name__ritemsrparseloggerwarning)rargskwargsmissingdriftr"exists installedtestedinstalled_base tested_basespecsoutdatedminimum_detailss` rrzHfExporter.validate_environment:s  R)) EC 5c$ O FIs#))--c2F!i5&8!*a!8!;$ll3215 ![0LL#~{!CD E IIipET (;(;'<?  rcDtt|jd)u Export the model and return the backend-specific program object. Args: model ([`PreTrainedModel`]): The model to export. sample_inputs (`dict[str, torch.Tensor | Cache]`): **Forward** kwargs — what you'd pass to `model(**sample_inputs)`. These are used directly as the example inputs during tracing. For an autoregressive decode-step export, this means you need to include `past_key_values`, `cache_position`, etc. If you only have generation-style inputs, use [`~HfExporter.export_for_generation`] instead — it runs `model.generate` for you and exports each stage. config ([`~transformers.exporters.configs.ExportConfigMixin`]): Backend-specific configuration. Returns: Backend-specific export artifact. z does not implement `export`. Pick a concrete exporter (`DynamoExporter`, `OnnxExporter`, `ExecutorchExporter`), or override `export` in your subclass with a backend-specific tracing pipeline that consumes `config` and returns the runtime artifact.)NotImplementedErrorr.r/)rmodel sample_inputsconfigs rexportzHfExporter.exportds+2"Dz""#$0 0  rc  t||}t|tr@t|t|z }|r$t dt |dt |d|}ntj ||}i}|jD]#\}\} } |j| | ||||<%|S#t$rL} tt|jd|dt| jdt| d| d } ~ wwxYw) u Decompose a generative model and export each component independently. Thin wrapper around [`~exporters.utils.decompose_for_generation`] that calls [`~HfExporter.export`] on every returned `(submodel, forward_inputs)` pair. If you need the intermediate `(submodel, forward_inputs)` pairs (for verification, custom inputs, skipping a stage, …), call [`~exporters.utils.decompose_for_generation`] directly. Args: model ([`PreTrainedModel`]): The generative model to export. Must support `model.generate(**sample_inputs)`. sample_inputs (`dict[str, torch.Tensor | Cache]`): **Generate** kwargs — what you'd pass to `model.generate(**sample_inputs)` (typically `input_ids` + `attention_mask`, plus any modality inputs like `pixel_values` / `input_features` for multi-modal models). Per-stage forward kwargs are captured internally. config ([`~transformers.exporters.configs.ExportConfigMixin`] or `dict[str, ExportConfigMixin]`): Backend-specific configuration. Pass a single config to apply to every component, or a `dict` keyed by component name (e.g. `"image_encoder"`, `"language_model"`, `"lm_head"`, `"decode"`) to override per-component — all component names must be present in the dict. Returns: `dict[str, Any]`: `{component_name: backend_specific_artifact}` — same keys as [`~exporters.utils.decompose_for_generation`]. Values are whatever [`~HfExporter.export`] returns for the concrete backend (`ExportedProgram`, `ONNXProgram`, `ExecutorchProgramManager`). z4Per-component `config` dict is missing entries for: z$. Expected one entry per component: .)rFz.export failed on component 'z ' (submodel=z , input keys=z).N)r isinstancedictset ValueErrorsortedfromkeysr0rG Exception RuntimeErrorr.r/list) rrDrErF componentsr6configsexportednamesubmodel subinputses rexport_for_generationz HfExporter.export_for_generations/D.e]C fd #*oF 3G J6RY?J[\99? 9K8LAOGmmJ7G&(+5+;+;+=  'D'8Y !%XyQU!W   "Dz**++HO!!%h!8!8 9tIFWWY[ sB-- D6AC==DN)rDrrE)MutableMapping[str, torch.Tensor | Cache]rFr)rDrrEr[rFz0ExportConfigMixin | dict[str, ExportConfigMixin]returnzdict[str, object]) r/ __module__ __qualname____doc__r__annotations__rrrrrrGrZr&rrrr*s $&y%#%L.%&(O^($(T  A "   >66A6A 6  6rr)r_ __future__rabcrrcollections.abcrtypingr packagingrutilsr utils.import_utilsr r rTrr get_loggerr/r2torch cache_utilsrmodeling_utilsrrr&rrrls[:"#* J&+   H %'4PPr