gguf : fix resource leaks (ggerganov#6061)

There several places where a gguf context is allocated. A call to gguf_free
is missing in some error paths. Also on linux, llama-bench was missing a
fclose.

examples/gguf/gguf.cpp

@@ -211,6 +211,7 @@ static bool gguf_ex_read_1(const std::string & fname) {
  for (int j = 0; j < ggml_nelements(cur); ++j) {
  if (data[j] != 100 + i) {
  fprintf(stderr, "%s: tensor[%d]: data[%d] = %f\n", __func__, i, j, data[j]);
+ gguf_free(ctx);
  return false;
  }
  }

examples/llama-bench/llama-bench.cpp

@@ -103,6 +103,7 @@ static std::string get_cpu_info() {
  }
  }
  }
+ fclose(f);
  }
 #endif
  // TODO: other platforms

examples/llava/clip.cpp

@@ -995,13 +995,15 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
  if (!new_clip->ctx_data) {
  fprintf(stderr, "%s: ggml_init() failed\n", __func__);
  clip_free(new_clip);
+ gguf_free(ctx);
  return nullptr;
  }
 
  auto fin = std::ifstream(fname, std::ios::binary);
  if (!fin) {
  printf("cannot open model file for loading tensors\n");
  clip_free(new_clip);
+ gguf_free(ctx);
  return nullptr;
  }
 
@@ -1023,6 +1025,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
  if (!fin) {
  printf("%s: failed to seek for tensor %s\n", __func__, name);
  clip_free(new_clip);
+ gguf_free(ctx);
  return nullptr;
  }
  int num_bytes = ggml_nbytes(cur);
@@ -1908,6 +1911,7 @@ bool clip_model_quantize(const char * fname_inp, const char * fname_out, const i
  break;
  default:
  printf("Please use an input file in f32 or f16\n");
+ gguf_free(ctx_out);
  return false;
  }
 

examples/train-text-from-scratch/train-text-from-scratch.cpp

@@ -711,6 +711,7 @@ static bool load_checkpoint_file(const char * filename, struct my_llama_model *
 
  load_checkpoint_gguf(fctx, f_ggml_ctx, model, train);
 
+ gguf_free(fctx);
  return true;
 }