Python · 1325 bytes Raw Blame History
1 """GGUF export pipeline — convert trained adapters to Ollama-consumable files.
2
3 Heavy imports (`torch`, `peft`,
4 `transformers`) stay deferred; subprocess calls to the vendored
5 `llama.cpp` tools go through `dlm.export.quantize.run_checked`.
6 """
7
8 from __future__ import annotations
9
10 from dlm.export.errors import (
11 ExportError,
12 ExportManifestError,
13 PreflightError,
14 SubprocessError,
15 UnknownExportTargetError,
16 UnsafeMergeError,
17 VendoringError,
18 )
19 from dlm.export.manifest import (
20 EXPORT_MANIFEST_FILENAME,
21 ExportArtifact,
22 ExportManifest,
23 load_export_manifest,
24 save_export_manifest,
25 )
26 from dlm.export.plan import (
27 DEFAULT_QUANT,
28 QUANT_BYTES_PER_PARAM,
29 ExportPlan,
30 QuantLevel,
31 resolve_export_plan,
32 valid_quants,
33 )
34 from dlm.export.runner import ExportResult, run_export
35
36 __all__ = [
37 "DEFAULT_QUANT",
38 "EXPORT_MANIFEST_FILENAME",
39 "ExportArtifact",
40 "ExportError",
41 "ExportManifest",
42 "ExportManifestError",
43 "ExportPlan",
44 "ExportResult",
45 "PreflightError",
46 "QUANT_BYTES_PER_PARAM",
47 "QuantLevel",
48 "SubprocessError",
49 "UnknownExportTargetError",
50 "UnsafeMergeError",
51 "VendoringError",
52 "load_export_manifest",
53 "resolve_export_plan",
54 "run_export",
55 "save_export_manifest",
56 "valid_quants",
57 ]