instructSAM.cpp/src/instructsam_lm_runner.cpp at main · smellslikeml/instructSAM.cpp

Repository navigation

#include "sam3/instructsam_lm_runner.h"

#include <stdexcept>

#include <string>

namespace sam3 {

// Skeleton — no llama.cpp linkage yet. See LM_INTEGRATION_PLAN.md for

// the concrete API-surface investigation and 5-day roadmap. Implementing

// this requires linking libllama + libmtmd from the llama.cpp build.

// This file exists so downstream code (CLI orchestrator, etc.) can

// compile against InstructsamLmRunner's interface while the impl grows.

struct InstructsamLmRunner::Impl {

std::string lm_gguf_path;

std::string mmproj_path;

std::string grounding_gguf;

int max_generated_tokens;

};

InstructsamLmRunner::InstructsamLmRunner(

const std::string & lm_gguf_path,

const std::string & mmproj_path,

const std::string & grounding_gguf,

int max_generated_tokens

) : impl_(std::make_unique<Impl>()) {

impl_->lm_gguf_path = lm_gguf_path;

impl_->mmproj_path = mmproj_path;

impl_->grounding_gguf = grounding_gguf;

impl_->max_generated_tokens = max_generated_tokens;

}

InstructsamLmRunner::~InstructsamLmRunner() = default;

LmOutput InstructsamLmRunner::run(const std::string & /*image_path*/,

const std::string & /*query*/) {

throw std::runtime_error(

"InstructsamLmRunner::run() is a skeleton — impl deferred to "

"Piece 3 Day 1-5 (see docs/instructsam/LM_INTEGRATION_PLAN.md). "

"Requires linking libllama + libmtmd from llama.cpp.");

}

} // namespace sam3