Skip to content

Commit ee681bf

Browse files
Jack-Khuumalfet
Jack-Khuu
authored and
malfet
committed
Remove args from help that we don't wish to advertise yet (#912)
1 parent 0f68cca commit ee681bf

File tree

1 file changed

+120
-91
lines changed

1 file changed

+120
-91
lines changed

‎cli.py

Lines changed: 120 additions & 91 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,7 @@
44
# This source code is licensed under the license found in the
55
# LICENSE file in the root directory of this source tree.
66

7+
import argparse
78
import json
89
import logging
910
import os
@@ -43,44 +44,46 @@ def check_args(args, verb: str) -> None:
4344
download_and_convert(args.model, args.model_directory, args.hf_token)
4445

4546

47+
# Given a arg parser and a subcommand (verb), add the appropriate arguments
48+
# for that subcommand.
4649
def add_arguments_for_verb(parser, verb: str) -> None:
47-
# Model specification. TODO Simplify this.
48-
# A model can be specified using a positional model name or HuggingFace
49-
# path. Alternatively, the model can be specified via --gguf-path or via
50-
# an explicit --checkpoint-dir, --checkpoint-path, or --tokenizer-path.
51-
50+
# Argument closure for inventory related subcommands
5251
if verb in INVENTORY_VERBS:
5352
_configure_artifact_inventory_args(parser, verb)
5453
_add_cli_metadata_args(parser)
5554
return
5655

56+
# Model specification
57+
# A model can be specified using a positional model name or checkpoint path
5758
parser.add_argument(
5859
"model",
5960
type=str,
6061
nargs="?",
6162
default=None,
6263
help="Model name for well-known models",
6364
)
65+
parser.add_argument(
66+
"--checkpoint-path",
67+
type=Path,
68+
default="not_specified",
69+
help="Use the specified model checkpoint path",
70+
)
6471

72+
# Add thematic argument groups based on the subcommand
6573
if verb in ["browser", "chat", "generate"]:
6674
_add_generation_args(parser)
75+
if verb == "eval":
76+
_add_evaluation_args(parser)
77+
78+
# Add argument groups for exported model path IO
79+
_add_exported_input_path_args(parser)
80+
_add_export_output_path_args(parser)
6781

68-
parser.add_argument(
69-
"--distributed",
70-
action="store_true",
71-
help="Whether to enable distributed inference",
72-
)
7382
parser.add_argument(
7483
"--is-chat-model",
7584
action="store_true",
7685
help="Indicate that the model was trained to support chat functionality",
7786
)
78-
parser.add_argument(
79-
"--seed",
80-
type=int,
81-
default=None,
82-
help="Initialize torch seed",
83-
)
8487
parser.add_argument(
8588
"--compile",
8689
action="store_true",
@@ -91,52 +94,6 @@ def add_arguments_for_verb(parser, verb: str) -> None:
9194
action="store_true",
9295
help="Whether to compile the prefill. Improves prefill perf, but has higher compile times.",
9396
)
94-
parser.add_argument(
95-
"--profile",
96-
type=Path,
97-
default=None,
98-
help="Profile path.",
99-
)
100-
parser.add_argument(
101-
"--draft-checkpoint-path",
102-
type=Path,
103-
default=None,
104-
help="Use the specified draft checkpoint path",
105-
)
106-
parser.add_argument(
107-
"--checkpoint-path",
108-
type=Path,
109-
default="not_specified",
110-
help="Use the specified model checkpoint path",
111-
)
112-
parser.add_argument(
113-
"--dcp-dir",
114-
type=Path,
115-
default=None,
116-
help="Use the specified model checkpoint directory",
117-
)
118-
parser.add_argument(
119-
"--params-path",
120-
type=Path,
121-
default=None,
122-
help="Use the specified parameter file",
123-
)
124-
parser.add_argument(
125-
"--gguf-path",
126-
type=Path,
127-
default=None,
128-
help="Use the specified GGUF model file",
129-
)
130-
parser.add_argument(
131-
"--tokenizer-path",
132-
type=Path,
133-
default=None,
134-
help="Use the specified model tokenizer file",
135-
)
136-
137-
_add_exported_model_input_args(parser)
138-
_add_export_output_path_args(parser)
139-
14097
parser.add_argument(
14198
"--dtype",
14299
default="fast",
@@ -152,34 +109,13 @@ def add_arguments_for_verb(parser, verb: str) -> None:
152109
+ "modes are: embedding, linear:int8, linear:int4, linear:a8w4dq, precision."
153110
),
154111
)
155-
parser.add_argument(
156-
"--draft-quantize",
157-
type=str,
158-
default="{ }",
159-
help=(
160-
"Quantization options. Same format as quantize, "
161-
+ "or 'quantize' to indicate same options specified by "
162-
+ "--quantize to main model. Applied to draft model."
163-
),
164-
)
165-
parser.add_argument(
166-
"--params-table",
167-
type=str,
168-
default=None,
169-
choices=allowable_params_table(),
170-
help="Parameter table to use",
171-
)
172112
parser.add_argument(
173113
"--device",
174114
type=str,
175115
default=default_device,
176116
choices=["fast", "cpu", "cuda", "mps"],
177117
help="Hardware device to use. Options: cpu, cuda, mps",
178118
)
179-
180-
if verb == "eval":
181-
_add_evaluation_args(parser)
182-
183119
parser.add_argument(
184120
"--hf-token",
185121
type=str,
@@ -192,6 +128,12 @@ def add_arguments_for_verb(parser, verb: str) -> None:
192128
default=default_model_dir,
193129
help=f"The directory to store downloaded model artifacts. Default: {default_model_dir}",
194130
)
131+
parser.add_argument(
132+
"--profile",
133+
type=Path,
134+
default=None,
135+
help="Profile path.",
136+
)
195137
parser.add_argument(
196138
"--port",
197139
type=int,
@@ -200,6 +142,11 @@ def add_arguments_for_verb(parser, verb: str) -> None:
200142
)
201143
_add_cli_metadata_args(parser)
202144

145+
# WIP Features (suppressed from --help)
146+
_add_distributed_args(parser)
147+
_add_custom_model_args(parser)
148+
_add_speculative_execution_args(parser)
149+
203150

204151
# Add CLI Args representing user provided exported model files
205152
def _add_export_output_path_args(parser) -> None:
@@ -219,7 +166,7 @@ def _add_export_output_path_args(parser) -> None:
219166

220167

221168
# Add CLI Args representing user provided exported model files
222-
def _add_exported_model_input_args(parser) -> None:
169+
def _add_exported_input_path_args(parser) -> None:
223170
exported_model_path_parser = parser.add_argument_group("Exported Model Path Args", "Specify the path of the exported model files to ingest")
224171
exported_model_path_parser.add_argument(
225172
"--dso-path",
@@ -235,14 +182,20 @@ def _add_exported_model_input_args(parser) -> None:
235182
)
236183

237184

238-
# Add CLI Args that are relevant to any subcommand execution
185+
# Add CLI Args that are general to subcommand cli execution
239186
def _add_cli_metadata_args(parser) -> None:
240187
parser.add_argument(
241188
"-v",
242189
"--verbose",
243190
action="store_true",
244191
help="Verbose output",
245192
)
193+
parser.add_argument(
194+
"--seed",
195+
type=int,
196+
default=None,
197+
help="Initialize torch seed",
198+
)
246199

247200

248201
# Configure CLI Args specific to Model Artifact Management
@@ -318,12 +271,6 @@ def _add_generation_args(parser) -> None:
318271
action="store_true",
319272
help="Whether to perform prefill sequentially. Only used for model debug.",
320273
)
321-
generator_parser.add_argument(
322-
"--speculate-k",
323-
type=int,
324-
default=5,
325-
help="Speculative execution depth",
326-
)
327274

328275

329276
# Add CLI Args specific to Model Evaluation
@@ -350,6 +297,88 @@ def _add_evaluation_args(parser) -> None:
350297
)
351298

352299

300+
# Add CLI Args related to distributed inference
301+
# This feature is currently a [WIP] and hidden from --help
302+
def _add_distributed_args(parser) -> None:
303+
parser.add_argument(
304+
"--distributed",
305+
action="store_true",
306+
help=argparse.SUPPRESS,
307+
# "Whether to enable distributed inference",
308+
)
309+
parser.add_argument(
310+
"--dcp-dir",
311+
type=Path,
312+
default=None,
313+
help=argparse.SUPPRESS,
314+
# "Use the specified model checkpoint directory",
315+
)
316+
317+
318+
# Add CLI Args related to custom model inputs (e.g. GGUF)
319+
# This feature is currently a [WIP] and hidden from --help
320+
def _add_custom_model_args(parser) -> None:
321+
parser.add_argument(
322+
"--params-table",
323+
type=str,
324+
default=None,
325+
choices=allowable_params_table(),
326+
help=argparse.SUPPRESS,
327+
# "Parameter table to use",
328+
)
329+
parser.add_argument(
330+
"--params-path",
331+
type=Path,
332+
default=None,
333+
help=argparse.SUPPRESS,
334+
# "Use the specified parameter file",
335+
)
336+
parser.add_argument(
337+
"--gguf-path",
338+
type=Path,
339+
default=None,
340+
help=argparse.SUPPRESS,
341+
# "Use the specified GGUF model file",
342+
)
343+
parser.add_argument(
344+
"--tokenizer-path",
345+
type=Path,
346+
default=None,
347+
help=argparse.SUPPRESS,
348+
# "Use the specified model tokenizer file",
349+
)
350+
351+
352+
# Add CLI Args related to speculative execution
353+
# This feature is currently a [WIP] and hidden from --help
354+
def _add_speculative_execution_args(parser) -> None:
355+
parser.add_argument(
356+
"--speculate-k",
357+
type=int,
358+
default=5,
359+
help=argparse.SUPPRESS,
360+
# "Speculative execution depth",
361+
)
362+
parser.add_argument(
363+
"--draft-checkpoint-path",
364+
type=Path,
365+
default=None,
366+
help=argparse.SUPPRESS,
367+
# "Use the specified draft checkpoint path",
368+
)
369+
parser.add_argument(
370+
"--draft-quantize",
371+
type=str,
372+
default="{ }",
373+
help=argparse.SUPPRESS,
374+
# (
375+
# "Quantization options. Same format as quantize, "
376+
# + "or 'quantize' to indicate same options specified by "
377+
# + "--quantize to main model. Applied to draft model."
378+
# ),
379+
)
380+
381+
353382
def arg_init(args):
354383
if not (torch.__version__ > "2.3"):
355384
raise RuntimeError(

0 commit comments

Comments
 (0)