use crate::state::{HistoryEntry, LoadedModel, ModelRole, SessionState};
use entrenar_common::{EntrenarError, Result};
#[derive(Debug, Clone, PartialEq)]
pub enum Command {
Fetch { model_id: String, role: ModelRole },
Inspect { target: InspectTarget },
Memory {
batch_size: Option<u32>,
seq_len: Option<usize>,
},
Set { key: String, value: String },
Distill { dry_run: bool },
Export { format: String, path: String },
History,
Help { topic: Option<String> },
Clear,
Quit,
Unknown { input: String },
}
#[derive(Debug, Clone, PartialEq)]
pub enum InspectTarget {
Layers,
Memory,
All,
Model(String),
}
pub fn parse(input: &str) -> Result<Command> {
let input = input.trim();
if input.is_empty() {
return Ok(Command::Unknown {
input: String::new(),
});
}
let parts: Vec<&str> = input.split_whitespace().collect();
let cmd = parts[0].to_lowercase();
let args = &parts[1..];
match cmd.as_str() {
"fetch" | "download" => parse_fetch(args),
"inspect" | "show" => parse_inspect(args),
"memory" | "mem" => parse_memory(args),
"set" => parse_set(args),
"distill" | "train" => parse_distill(args),
"export" | "save" => parse_export(args),
"history" | "hist" => Ok(Command::History),
"help" | "?" => parse_help(args),
"clear" | "cls" => Ok(Command::Clear),
"quit" | "exit" | "q" => Ok(Command::Quit),
_ => Ok(Command::Unknown {
input: input.to_string(),
}),
}
}
fn parse_fetch(args: &[&str]) -> Result<Command> {
if args.is_empty() {
return Err(EntrenarError::ConfigValue {
field: "model_id".into(),
message: "No model ID provided".into(),
suggestion: "Usage: fetch <model_id> [--teacher|--student]".into(),
});
}
let model_id = args[0].to_string();
let role = if args.contains(&"--teacher") {
ModelRole::Teacher
} else if args.contains(&"--student") {
ModelRole::Student
} else {
ModelRole::None
};
Ok(Command::Fetch { model_id, role })
}
fn parse_inspect(args: &[&str]) -> Result<Command> {
let target = if args.is_empty() {
InspectTarget::All
} else {
match args[0].to_lowercase().as_str() {
"layers" | "layer" => InspectTarget::Layers,
"memory" | "mem" => InspectTarget::Memory,
"all" => InspectTarget::All,
name => InspectTarget::Model(name.to_string()),
}
};
Ok(Command::Inspect { target })
}
fn parse_memory(args: &[&str]) -> Result<Command> {
let mut batch_size = None;
let mut seq_len = None;
let mut i = 0;
while i < args.len() {
match args[i] {
"--batch" | "-b" if i + 1 < args.len() => {
batch_size = args[i + 1].parse().ok();
i += 2;
}
"--seq" | "-s" if i + 1 < args.len() => {
seq_len = args[i + 1].parse().ok();
i += 2;
}
_ => i += 1,
}
}
Ok(Command::Memory {
batch_size,
seq_len,
})
}
fn parse_set(args: &[&str]) -> Result<Command> {
if args.len() < 2 {
return Err(EntrenarError::ConfigValue {
field: "set".into(),
message: "Not enough arguments".into(),
suggestion: "Usage: set <key> <value>".into(),
});
}
Ok(Command::Set {
key: args[0].to_string(),
value: args[1..].join(" "),
})
}
fn parse_distill(args: &[&str]) -> Result<Command> {
let dry_run = args.contains(&"--dry-run") || args.contains(&"-n");
Ok(Command::Distill { dry_run })
}
fn parse_export(args: &[&str]) -> Result<Command> {
if args.len() < 2 {
return Err(EntrenarError::ConfigValue {
field: "export".into(),
message: "Not enough arguments".into(),
suggestion: "Usage: export <format> <path>".into(),
});
}
Ok(Command::Export {
format: args[0].to_string(),
path: args[1].to_string(),
})
}
fn parse_help(args: &[&str]) -> Result<Command> {
let topic = args.first().map(ToString::to_string);
Ok(Command::Help { topic })
}
pub fn execute(cmd: &Command, state: &mut SessionState) -> Result<String> {
let start = std::time::Instant::now();
let result = match cmd {
Command::Fetch { model_id, role } => execute_fetch(model_id, *role, state),
Command::Inspect { target } => execute_inspect(target, state),
Command::Memory {
batch_size,
seq_len,
} => execute_memory(*batch_size, *seq_len, state),
Command::Set { key, value } => execute_set(key, value, state),
Command::Distill { dry_run } => execute_distill(*dry_run, state),
Command::Export { format, path } => execute_export(format, path, state),
Command::History => execute_history(state),
Command::Help { topic } => execute_help(topic.as_deref()),
Command::Clear => Ok(String::new()),
Command::Quit => Ok("Goodbye!".to_string()),
Command::Unknown { input } => {
if input.is_empty() {
Ok(String::new())
} else {
Err(EntrenarError::ConfigValue {
field: "command".into(),
message: format!("Unknown command: {input}"),
suggestion: "Type 'help' for available commands".into(),
})
}
}
};
let duration_ms = start.elapsed().as_millis() as u64;
let success = result.is_ok();
if !matches!(
cmd,
Command::Help { .. }
| Command::History
| Command::Clear
| Command::Quit
| Command::Unknown { .. }
) {
let cmd_str = format!("{cmd:?}");
state.add_to_history(HistoryEntry::new(cmd_str, duration_ms, success));
state.record_command(duration_ms, success);
}
result
}
fn execute_fetch(model_id: &str, role: ModelRole, state: &mut SessionState) -> Result<String> {
let model = LoadedModel {
id: model_id.to_string(),
path: std::path::PathBuf::from(format!("/tmp/models/{}", model_id.replace('/', "_"))),
architecture: detect_architecture(model_id),
parameters: estimate_params(model_id),
layers: estimate_layers(model_id),
hidden_dim: 4096,
role,
};
let name = if role == ModelRole::Teacher {
"teacher"
} else if role == ModelRole::Student {
"student"
} else {
model_id.split('/').next_back().unwrap_or(model_id)
};
state.add_model(name.to_string(), model.clone());
Ok(format!(
"✓ Fetched {}\n Architecture: {}\n Parameters: {:.1}B\n Layers: {}",
model_id,
model.architecture,
model.parameters as f64 / 1e9,
model.layers
))
}
fn execute_inspect(target: &InspectTarget, state: &SessionState) -> Result<String> {
match target {
InspectTarget::All => {
if state.loaded_models().is_empty() {
return Ok("No models loaded. Use 'fetch <model_id>' to load a model.".to_string());
}
let mut output = String::from("Loaded Models:\n");
for (name, model) in state.loaded_models() {
output.push_str(&format!(
" {} ({}): {:.1}B params, {} layers\n",
name,
model.id,
model.parameters as f64 / 1e9,
model.layers
));
}
Ok(output)
}
InspectTarget::Layers => {
let mut output = String::from("Layer Analysis:\n");
for (name, model) in state.loaded_models() {
output.push_str(&format!(
" {}: {} layers, hidden_dim={}\n",
name, model.layers, model.hidden_dim
));
}
Ok(output)
}
InspectTarget::Memory => execute_memory(None, None, state),
InspectTarget::Model(name) => {
if let Some(model) = state.get_model(name) {
Ok(format!(
"Model: {}\n ID: {}\n Path: {}\n Architecture: {}\n Parameters: {:.1}B\n Layers: {}\n Hidden Dim: {}",
name, model.id, model.path.display(), model.architecture,
model.parameters as f64 / 1e9, model.layers, model.hidden_dim
))
} else {
Err(EntrenarError::ModelNotFound { path: name.into() })
}
}
}
}
fn execute_memory(
batch_size: Option<u32>,
seq_len: Option<usize>,
state: &SessionState,
) -> Result<String> {
let batch = batch_size.unwrap_or(state.preferences().default_batch_size);
let seq = seq_len.unwrap_or(state.preferences().default_seq_len);
let total_params: u64 = state.loaded_models().values().map(|m| m.parameters).sum();
let model_mem = total_params * 2; let activation_mem = u64::from(batch) * (seq as u64) * 4096 * 32 * 2;
let total = model_mem + activation_mem;
Ok(format!(
"Memory Estimate (batch={}, seq={}):\n Model: {:.1} GB\n Activations: {:.1} GB\n Total: {:.1} GB",
batch, seq,
model_mem as f64 / 1e9,
activation_mem as f64 / 1e9,
total as f64 / 1e9
))
}
fn execute_set(key: &str, value: &str, state: &mut SessionState) -> Result<String> {
match key {
"batch_size" | "batch" => {
let v: u32 = value.parse().map_err(|_| EntrenarError::ConfigValue {
field: "batch_size".into(),
message: "Invalid number".into(),
suggestion: "Use a positive integer".into(),
})?;
state.preferences_mut().default_batch_size = v;
Ok(format!("Set batch_size = {v}"))
}
"seq_len" | "seq" => {
let v: usize = value.parse().map_err(|_| EntrenarError::ConfigValue {
field: "seq_len".into(),
message: "Invalid number".into(),
suggestion: "Use a positive integer".into(),
})?;
state.preferences_mut().default_seq_len = v;
Ok(format!("Set seq_len = {v}"))
}
_ => Err(EntrenarError::ConfigValue {
field: key.into(),
message: "Unknown setting".into(),
suggestion: "Available settings: batch_size, seq_len".into(),
}),
}
}
fn execute_distill(dry_run: bool, state: &SessionState) -> Result<String> {
let teacher = state
.loaded_models()
.values()
.find(|m| m.role == ModelRole::Teacher);
let student = state
.loaded_models()
.values()
.find(|m| m.role == ModelRole::Student);
if teacher.is_none() {
return Err(EntrenarError::ConfigValue {
field: "teacher".into(),
message: "No teacher model loaded".into(),
suggestion: "Use 'fetch <model_id> --teacher' to load a teacher model".into(),
});
}
if student.is_none() {
return Err(EntrenarError::ConfigValue {
field: "student".into(),
message: "No student model loaded".into(),
suggestion: "Use 'fetch <model_id> --student' to load a student model".into(),
});
}
if dry_run {
let teacher = teacher.expect("teacher validated non-None above");
let student = student.expect("student validated non-None above");
Ok(format!(
"Dry run configuration:\n Teacher: {} ({:.1}B)\n Student: {} ({:.1}B)\n Ready to train",
teacher.id, teacher.parameters as f64 / 1e9,
student.id, student.parameters as f64 / 1e9
))
} else {
Ok("Training started... (simulated)".to_string())
}
}
fn execute_export(format: &str, path: &str, _state: &SessionState) -> Result<String> {
Ok(format!("Exported to {path} in {format} format"))
}
fn execute_history(state: &SessionState) -> Result<String> {
if state.history().is_empty() {
return Ok("No command history.".to_string());
}
let mut output = String::from("Command History:\n");
for (i, entry) in state.history().iter().enumerate() {
let status = if entry.success { "✓" } else { "✗" };
output.push_str(&format!(
" {}. {} {} ({}ms)\n",
i + 1,
status,
entry.command,
entry.duration_ms
));
}
Ok(output)
}
fn execute_help(topic: Option<&str>) -> Result<String> {
match topic {
Some("fetch") => Ok(
"fetch <model_id> [--teacher|--student]\n Download a model from HuggingFace"
.to_string(),
),
Some("inspect") => {
Ok("inspect [layers|memory|all|<model>]\n Inspect loaded models".to_string())
}
Some("memory") => {
Ok("memory [--batch <n>] [--seq <n>]\n Estimate memory requirements".to_string())
}
Some("distill") => Ok("distill [--dry-run]\n Run distillation training".to_string()),
_ => Ok("Available commands:
fetch <model> Download model from HuggingFace
inspect [target] Inspect loaded models
memory Estimate memory requirements
set <key> <value> Configure settings
distill Run distillation
export <fmt> <path> Export model
history Show command history
help [topic] Show help
quit Exit shell"
.to_string()),
}
}
const ARCH_PATTERNS: &[(&[&str], &str)] = &[
(&["qwen"], "qwen"),
(&["phi"], "phi"),
(&["falcon"], "falcon"),
(&["mistral", "mixtral"], "mistral"),
(&["llama"], "llama"),
(&["bert"], "bert"),
(&["gpt"], "gpt"),
];
fn detect_architecture(model_id: &str) -> String {
let lower = model_id.to_lowercase();
for (patterns, arch) in ARCH_PATTERNS {
if patterns.iter().any(|p| lower.contains(p)) {
return (*arch).to_string();
}
}
eprintln!(
"Warning: could not detect architecture from model ID '{model_id}', \
defaulting to 'unknown' (use tensor-based detection for accuracy)"
);
"unknown".to_string()
}
fn estimate_params(model_id: &str) -> u64 {
let lower = model_id.to_lowercase();
if lower.contains("70b") {
70_000_000_000
} else if lower.contains("13b") {
13_000_000_000
} else if lower.contains("7b") {
7_000_000_000
} else if lower.contains("1.1b") || lower.contains("1b") {
1_100_000_000
} else if lower.contains("base") {
350_000_000
} else {
1_000_000_000
}
}
fn estimate_layers(model_id: &str) -> u32 {
let lower = model_id.to_lowercase();
if lower.contains("70b") {
80
} else if lower.contains("13b") {
40
} else if lower.contains("7b") {
32
} else if lower.contains("base") {
12
} else {
24
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_parse_fetch() {
let cmd = parse("fetch meta-llama/Llama-2-7b --teacher").expect("parsing should succeed");
assert!(matches!(
cmd,
Command::Fetch {
role: ModelRole::Teacher,
..
}
));
let cmd =
parse("fetch TinyLlama/TinyLlama-1.1B --student").expect("parsing should succeed");
assert!(matches!(
cmd,
Command::Fetch {
role: ModelRole::Student,
..
}
));
}
#[test]
fn test_parse_inspect() {
assert!(matches!(
parse("inspect").expect("parsing should succeed"),
Command::Inspect {
target: InspectTarget::All
}
));
assert!(matches!(
parse("inspect layers").expect("parsing should succeed"),
Command::Inspect {
target: InspectTarget::Layers
}
));
}
#[test]
fn test_parse_memory() {
let cmd = parse("memory --batch 64 --seq 1024").expect("parsing should succeed");
if let Command::Memory {
batch_size,
seq_len,
} = cmd
{
assert_eq!(batch_size, Some(64));
assert_eq!(seq_len, Some(1024));
} else {
panic!("Expected Memory command");
}
}
#[test]
fn test_parse_quit_variants() {
assert!(matches!(
parse("quit").expect("parsing should succeed"),
Command::Quit
));
assert!(matches!(
parse("exit").expect("parsing should succeed"),
Command::Quit
));
assert!(matches!(
parse("q").expect("parsing should succeed"),
Command::Quit
));
}
#[test]
fn test_execute_fetch() {
let mut state = SessionState::new();
let result = execute_fetch("meta-llama/Llama-2-7b", ModelRole::Teacher, &mut state);
assert!(result.is_ok());
assert!(state.get_model("teacher").is_some());
}
#[test]
fn test_execute_set() {
let mut state = SessionState::new();
execute_set("batch_size", "64", &mut state).expect("operation should succeed");
assert_eq!(state.preferences().default_batch_size, 64);
execute_set("seq_len", "1024", &mut state).expect("operation should succeed");
assert_eq!(state.preferences().default_seq_len, 1024);
}
#[test]
fn test_unknown_command() {
let cmd = parse("foobar").expect("parsing should succeed");
assert!(matches!(cmd, Command::Unknown { .. }));
}
#[test]
fn test_parse_fetch_missing_model() {
let result = parse("fetch");
assert!(result.is_err());
}
#[test]
fn test_parse_set_not_enough_args() {
let result = parse("set batch_size");
assert!(result.is_err());
}
#[test]
fn test_parse_export_not_enough_args() {
let result = parse("export safetensors");
assert!(result.is_err());
}
#[test]
fn test_parse_export_valid() {
let cmd = parse("export safetensors /tmp/model.st").expect("parsing should succeed");
if let Command::Export { format, path } = cmd {
assert_eq!(format, "safetensors");
assert_eq!(path, "/tmp/model.st");
} else {
panic!("Expected Export command");
}
}
#[test]
fn test_parse_help_with_topic() {
let cmd = parse("help fetch").expect("parsing should succeed");
if let Command::Help { topic } = cmd {
assert_eq!(topic, Some("fetch".to_string()));
} else {
panic!("Expected Help command");
}
}
#[test]
fn test_parse_distill_dry_run() {
let cmd = parse("distill --dry-run").expect("parsing should succeed");
if let Command::Distill { dry_run } = cmd {
assert!(dry_run);
} else {
panic!("Expected Distill command");
}
}
#[test]
fn test_parse_distill_short_flag() {
let cmd = parse("distill -n").expect("parsing should succeed");
if let Command::Distill { dry_run } = cmd {
assert!(dry_run);
} else {
panic!("Expected Distill command");
}
}
#[test]
fn test_parse_inspect_model() {
let cmd = parse("inspect teacher").expect("parsing should succeed");
if let Command::Inspect { target } = cmd {
assert_eq!(target, InspectTarget::Model("teacher".to_string()));
} else {
panic!("Expected Inspect command");
}
}
#[test]
fn test_parse_inspect_memory() {
let cmd = parse("inspect memory").expect("parsing should succeed");
assert!(matches!(
cmd,
Command::Inspect {
target: InspectTarget::Memory
}
));
}
#[test]
fn test_parse_command_aliases() {
assert!(matches!(
parse("download model").expect("parsing should succeed"),
Command::Fetch { .. }
));
assert!(matches!(
parse("show layers").expect("parsing should succeed"),
Command::Inspect { .. }
));
assert!(matches!(
parse("mem").expect("parsing should succeed"),
Command::Memory { .. }
));
assert!(matches!(
parse("train").expect("parsing should succeed"),
Command::Distill { .. }
));
assert!(matches!(
parse("save gguf /tmp/out").expect("parsing should succeed"),
Command::Export { .. }
));
assert!(matches!(
parse("cls").expect("parsing should succeed"),
Command::Clear
));
assert!(matches!(
parse("?").expect("parsing should succeed"),
Command::Help { .. }
));
assert!(matches!(
parse("hist").expect("parsing should succeed"),
Command::History
));
}
#[test]
fn test_execute_inspect_no_models() {
let state = SessionState::new();
let result = execute_inspect(&InspectTarget::All, &state);
assert!(result
.expect("load should succeed")
.contains("No models loaded"));
}
#[test]
fn test_execute_inspect_layers() {
let mut state = SessionState::new();
let model = LoadedModel {
id: "test".to_string(),
path: std::path::PathBuf::from("/tmp"),
architecture: "llama".to_string(),
parameters: 7_000_000_000,
layers: 32,
hidden_dim: 4096,
role: ModelRole::None,
};
state.add_model("test".to_string(), model);
let result =
execute_inspect(&InspectTarget::Layers, &state).expect("operation should succeed");
assert!(result.contains("32 layers"));
}
#[test]
fn test_execute_inspect_model_not_found() {
let state = SessionState::new();
let result = execute_inspect(&InspectTarget::Model("unknown".to_string()), &state);
assert!(result.is_err());
}
#[test]
fn test_execute_history_empty() {
let state = SessionState::new();
let result = execute_history(&state).expect("operation should succeed");
assert!(result.contains("No command history"));
}
#[test]
fn test_execute_help_topics() {
let fetch_help = execute_help(Some("fetch")).expect("operation should succeed");
assert!(fetch_help.contains("Download"));
let inspect_help = execute_help(Some("inspect")).expect("operation should succeed");
assert!(inspect_help.contains("Inspect"));
let memory_help = execute_help(Some("memory")).expect("operation should succeed");
assert!(memory_help.contains("memory"));
let distill_help = execute_help(Some("distill")).expect("operation should succeed");
assert!(distill_help.contains("distill"));
let general_help = execute_help(None).expect("operation should succeed");
assert!(general_help.contains("Available commands"));
}
#[test]
fn test_detect_architecture_variants() {
assert_eq!(detect_architecture("meta-llama/Llama-2-7b"), "llama");
assert_eq!(detect_architecture("bert-base-uncased"), "bert");
assert_eq!(detect_architecture("openai-gpt"), "gpt");
assert_eq!(detect_architecture("mistralai/Mistral-7B"), "mistral");
assert_eq!(detect_architecture("Qwen/Qwen2.5-Coder-0.5B"), "qwen");
assert_eq!(detect_architecture("microsoft/phi-2"), "phi");
assert_eq!(detect_architecture("tiiuae/falcon-7b"), "falcon");
assert_eq!(detect_architecture("mistralai/Mixtral-8x7B"), "mistral");
assert_eq!(detect_architecture("custom-model"), "unknown");
}
#[test]
fn test_falsify_n04_mistral_before_llama_in_ambiguous_id() {
assert_eq!(
detect_architecture("mistral-llama-variant"),
"mistral",
"Mistral should take priority over LLaMA in ambiguous IDs"
);
}
#[test]
fn test_falsify_n04_unknown_is_explicit() {
let result = detect_architecture("totally-custom-model-v3");
assert_eq!(result, "unknown");
assert!(!result.is_empty());
}
#[test]
fn test_estimate_params_variants() {
assert_eq!(estimate_params("model-70b"), 70_000_000_000);
assert_eq!(estimate_params("model-13b"), 13_000_000_000);
assert_eq!(estimate_params("model-7b"), 7_000_000_000);
assert_eq!(estimate_params("model-1.1b"), 1_100_000_000);
assert_eq!(estimate_params("bert-base"), 350_000_000);
}
#[test]
fn test_estimate_layers_variants() {
assert_eq!(estimate_layers("model-70b"), 80);
assert_eq!(estimate_layers("model-13b"), 40);
assert_eq!(estimate_layers("model-7b"), 32);
assert_eq!(estimate_layers("bert-base"), 12);
}
#[test]
fn test_execute_set_invalid_number() {
let mut state = SessionState::new();
let result = execute_set("batch_size", "not_a_number", &mut state);
assert!(result.is_err());
}
#[test]
fn test_execute_set_unknown_key() {
let mut state = SessionState::new();
let result = execute_set("unknown_setting", "value", &mut state);
assert!(result.is_err());
}
#[test]
fn test_execute_distill_no_teacher() {
let state = SessionState::new();
let result = execute_distill(true, &state);
assert!(result.is_err());
}
#[test]
fn test_execute_distill_no_student() {
let mut state = SessionState::new();
let model = LoadedModel {
id: "teacher".to_string(),
path: std::path::PathBuf::from("/tmp"),
architecture: "llama".to_string(),
parameters: 7_000_000_000,
layers: 32,
hidden_dim: 4096,
role: ModelRole::Teacher,
};
state.add_model("teacher".to_string(), model);
let result = execute_distill(true, &state);
assert!(result.is_err());
}
#[test]
fn test_execute_distill_success() {
let mut state = SessionState::new();
let teacher = LoadedModel {
id: "teacher/model".to_string(),
path: std::path::PathBuf::from("/tmp/t"),
architecture: "llama".to_string(),
parameters: 7_000_000_000,
layers: 32,
hidden_dim: 4096,
role: ModelRole::Teacher,
};
state.add_model("teacher".to_string(), teacher);
let student = LoadedModel {
id: "student/model".to_string(),
path: std::path::PathBuf::from("/tmp/s"),
architecture: "llama".to_string(),
parameters: 1_000_000_000,
layers: 12,
hidden_dim: 2048,
role: ModelRole::Student,
};
state.add_model("student".to_string(), student);
let result = execute_distill(true, &state).expect("operation should succeed");
assert!(result.contains("Dry run"));
}
#[test]
fn test_execute_export() {
let state = SessionState::new();
let result = execute_export("safetensors", "/tmp/model.st", &state)
.expect("operation should succeed");
assert!(result.contains("Exported"));
}
#[test]
fn test_parse_empty_input() {
let cmd = parse("").expect("parsing should succeed");
assert!(matches!(cmd, Command::Unknown { .. }));
}
#[test]
fn test_execute_memory_with_args() {
let state = SessionState::new();
let result =
execute_memory(Some(64), Some(1024), &state).expect("operation should succeed");
assert!(result.contains("batch=64"));
assert!(result.contains("seq=1024"));
}
#[test]
fn test_command_enum_equality() {
assert_eq!(Command::Quit, Command::Quit);
assert_eq!(Command::Clear, Command::Clear);
assert_eq!(Command::History, Command::History);
}
}