mesh_llm_cli/
models.rs

1use clap::{Subcommand, ValueEnum};
2
3#[derive(Clone, Copy, Debug, Eq, PartialEq, ValueEnum)]
4pub enum ModelSearchSort {
5    Trending,
6    Downloads,
7    Likes,
8    Created,
9    Updated,
10    #[value(name = "parameters-desc", alias = "most-parameters")]
11    ParametersDesc,
12    #[value(name = "parameters-asc", alias = "least-parameters")]
13    ParametersAsc,
14}
15
16#[derive(Subcommand, Debug)]
17// CLI enums mirror clap's argument shape; boxing these fields would make the parser harder to maintain.
18#[allow(clippy::large_enum_variant)]
19pub enum ModelsCommand {
20    /// Package a GGUF model for distributed inference by splitting it into layer files on Hugging Face Jobs.
21    Package {
22        /// Source Hugging Face model ref (e.g. unsloth/Qwen3-235B-A22B-GGUF:UD-Q4_K_XL).
23        source_repo: Option<String>,
24        /// Quantization variant (deprecated; prefer source refs like repo:Q4_K_M).
25        #[arg(long)]
26        quant: Option<String>,
27        /// Target repo for the layer package (auto-derived if omitted).
28        #[arg(long)]
29        target: Option<String>,
30        /// Override model ID in the manifest.
31        #[arg(long)]
32        model_id: Option<String>,
33        /// HF Job hardware flavor. Use auto for the default CPU splitter baseline.
34        #[arg(long, default_value = "auto")]
35        flavor: String,
36        /// Requested job timeout; raised automatically by model-size minimums.
37        #[arg(long, default_value = "1h")]
38        timeout: String,
39        /// Branch or tag of mesh-llm to build in the job.
40        #[arg(long, default_value = "main")]
41        mesh_llm_ref: String,
42        /// Publish a public package marked experimental and open an unmerged HF catalog PR.
43        #[arg(long)]
44        experimental: bool,
45        /// Explicitly keep this as a dry run. This is the default unless --confirm is set.
46        #[arg(long)]
47        dry_run: bool,
48        /// Actually submit the HF Job. Without this, the command only prints plan, spec, and max cost.
49        #[arg(long)]
50        confirm: bool,
51        /// Stream job logs after submission until completion.
52        #[arg(long)]
53        follow: bool,
54        /// Check status of a previously submitted job.
55        #[arg(long)]
56        status: Option<String>,
57        /// Fetch logs for a previously submitted job.
58        #[arg(long)]
59        logs: Option<String>,
60        /// Cancel a running job.
61        #[arg(long)]
62        cancel: Option<String>,
63        /// List recent package jobs.
64        #[arg(long)]
65        list: bool,
66        /// Upload the latest job script to the meshllm bucket (requires org access).
67        #[arg(long)]
68        update_script: bool,
69        /// Emit JSON output.
70        #[arg(long)]
71        json: bool,
72    },
73    /// List recommended models from the remote meshllm/catalog.
74    Recommended {
75        /// Emit JSON output.
76        #[arg(long)]
77        json: bool,
78    },
79    /// List installed local models from the HF cache.
80    Installed {
81        /// Emit JSON output.
82        #[arg(long)]
83        json: bool,
84    },
85    /// Preview or remove mesh-managed models from the Hugging Face cache.
86    Cleanup {
87        /// Only include models that mesh-llm has not used for the given age (for example 30d or 12h).
88        #[arg(long)]
89        unused_since: Option<String>,
90        /// Remove the selected files instead of printing a dry run preview.
91        #[arg(long)]
92        yes: bool,
93        /// Emit JSON output.
94        #[arg(long)]
95        json: bool,
96    },
97    /// Remove stale derived skippy stage artifacts from the mesh cache.
98    Prune {
99        /// Remove files instead of printing a dry run note.
100        #[arg(long)]
101        yes: bool,
102        /// Emit JSON output.
103        #[arg(long)]
104        json: bool,
105    },
106    /// Certify a Skippy layer package can be resolved, verified, and smoke-tested.
107    Certify {
108        /// Exact layer package ref, local package dir, or catalog model ref with a package mapping.
109        model: String,
110        /// Write the JSON certification report to this path.
111        #[arg(long)]
112        report_out: Option<std::path::PathBuf>,
113        /// Emit JSON output.
114        #[arg(long)]
115        json: bool,
116        /// Stop after package resolution, integrity checks, and local stage materialization.
117        #[arg(long)]
118        package_only: bool,
119        /// Existing mesh-llm OpenAI-compatible API base for runtime smoke gates.
120        #[arg(long)]
121        api_base: Option<String>,
122        /// Prompt for runtime smoke gates.
123        #[arg(long, default_value = "Say ok.")]
124        prompt: String,
125        /// Maximum tokens for runtime smoke gates.
126        #[arg(long, default_value_t = 2)]
127        max_tokens: u32,
128    },
129    // Delete variant defined with explicit clap args later in file (existing block).
130    /// List remote catalog models.
131    #[command(hide = true)]
132    List {
133        /// Emit JSON output.
134        #[arg(long)]
135        json: bool,
136    },
137    /// Search for catalog models and downloadable GGUF/MLX artifacts on Hugging Face.
138    Search {
139        /// Search terms.
140        #[arg(required = true)]
141        query: Vec<String>,
142        /// Filter search results to GGUF artifacts (default).
143        #[arg(long, conflicts_with = "mlx")]
144        gguf: bool,
145        /// Filter search results to MLX artifacts.
146        #[arg(long, conflicts_with = "gguf")]
147        mlx: bool,
148        /// Search only the remote meshllm/catalog.
149        #[arg(long)]
150        catalog: bool,
151        /// Maximum number of results to show.
152        #[arg(long, default_value = "20")]
153        limit: usize,
154        /// Sort search results.
155        #[arg(long, value_enum, default_value = "trending")]
156        sort: ModelSearchSort,
157        /// Emit JSON output.
158        #[arg(long)]
159        json: bool,
160    },
161    /// Show details for one exact model reference.
162    Show {
163        /// Exact remote catalog id or Hugging Face ref.
164        model: String,
165        /// Emit JSON output.
166        #[arg(long)]
167        json: bool,
168    },
169    /// Download one exact model reference.
170    Download {
171        /// Exact remote catalog id or Hugging Face ref.
172        model: String,
173        /// Also download the recommended draft model for speculative decoding.
174        #[arg(long)]
175        draft: bool,
176        /// Download the exact Hugging Face file directly, bypassing catalog layer-package resolution.
177        #[arg(long)]
178        direct: bool,
179        /// Emit JSON output.
180        #[arg(long)]
181        json: bool,
182    },
183    /// Check or refresh cached Hugging Face repos.
184    #[command(visible_alias = "update")]
185    Updates {
186        /// Repo id like Qwen/Qwen3-8B-GGUF.
187        repo: Option<String>,
188        /// Operate on every cached Hugging Face repo.
189        #[arg(long)]
190        all: bool,
191        /// Check for newer upstream revisions without refreshing local cache.
192        #[arg(long)]
193        check: bool,
194        /// Emit JSON output.
195        #[arg(long)]
196        json: bool,
197    },
198    /// Delete a specific model from local storage.
199    Delete {
200        /// Installed model stem or Hugging Face ref (e.g. `Qwen3.5-9B-BF16`, `org/repo`, or `org/repo:BF16`).
201        #[arg(required = true)]
202        model: String,
203        /// Skip dry-run preview and delete immediately.
204        #[arg(long)]
205        yes: bool,
206        /// Emit JSON output.
207        #[arg(long)]
208        json: bool,
209    },
210}
mesh_llm_cli/models.rs

mesh_llm_cli/
models.rs
