308 lines
11 KiB
C
308 lines
11 KiB
C
/* SPDX-License-Identifier: AGPL-3.0-or-later */
|
|
/* Copyright (C) 2026 David (Corey) Herrera */
|
|
/*
|
|
* test_multi_fact.c — HDC Memory Viability Test 5
|
|
*
|
|
* Multi-Fact Battery
|
|
*
|
|
* Method:
|
|
* 10 facts, each with matching question.
|
|
* For each fact: A/B test (no HM vs HM), cache_size=8.
|
|
* Binary scoring: output contains specific detail?
|
|
* Summary: X/10 improved, Y/10 same, Z/10 worse.
|
|
*
|
|
* Usage:
|
|
* ./examples/test_multi_fact \
|
|
* --model ../models/base/RWKV-X-3.6B-64k-Base.safetensors \
|
|
* --vocab rwkvx/data/rwkv_vocab_v20230424.txt \
|
|
* --filler ../docs/architecture/rwkv-x-deep-dive.md
|
|
*/
|
|
|
|
#include "rwkvx/rwkvx.h"
|
|
#include "tensor/tensor.h"
|
|
|
|
#include <ctype.h>
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include <time.h>
|
|
|
|
typedef struct {
|
|
const char *fact;
|
|
const char *question;
|
|
const char *key_detail; /* substring to check in output */
|
|
} fact_entry;
|
|
|
|
static const fact_entry FACTS[] = {
|
|
{
|
|
"The secret project codename is AURORA-7 and the lead researcher is Dr. Kimura.",
|
|
"What is the secret project codename and who leads it?",
|
|
"AURORA"
|
|
},
|
|
{
|
|
"The server room temperature must be maintained at exactly 18.5 degrees Celsius.",
|
|
"What temperature must the server room be maintained at?",
|
|
"18.5"
|
|
},
|
|
{
|
|
"Employee badge number 4472 belongs to Sarah Chen from the quantum computing division.",
|
|
"What is Sarah Chen's badge number?",
|
|
"4472"
|
|
},
|
|
{
|
|
"The annual budget for Project Helios is 2.7 million dollars, allocated by Director Williams.",
|
|
"What is the annual budget for Project Helios?",
|
|
"2.7"
|
|
},
|
|
{
|
|
"The backup protocol requires data replication to three separate data centers: Tokyo, Frankfurt, and Sao Paulo.",
|
|
"Which three data centers are used for backup replication?",
|
|
"Tokyo"
|
|
},
|
|
{
|
|
"The new encryption standard uses a 384-bit key length with the codename Obsidian Shield.",
|
|
"What key length does the Obsidian Shield encryption use?",
|
|
"384"
|
|
},
|
|
{
|
|
"Dr. Patel discovered that compound X-891 reduces neural inflammation by 67 percent in clinical trials.",
|
|
"What percentage does compound X-891 reduce neural inflammation?",
|
|
"67"
|
|
},
|
|
{
|
|
"The satellite launch window for Mission Vanguard is March 15 to March 22, from Pad 39B.",
|
|
"When is the launch window for Mission Vanguard?",
|
|
"March"
|
|
},
|
|
{
|
|
"The quantum processor has 127 qubits and operates at 15 millikelvin, manufactured by Rigetti.",
|
|
"How many qubits does the quantum processor have?",
|
|
"127"
|
|
},
|
|
{
|
|
"The ancient manuscript, catalogued as MS-7749, was found in a monastery near Thessaloniki in 1987.",
|
|
"Where was manuscript MS-7749 found?",
|
|
"Thessaloniki"
|
|
}
|
|
};
|
|
#define N_FACTS 10
|
|
|
|
static char* read_file(const char *path) {
|
|
FILE *fp = fopen(path, "r");
|
|
if (!fp) return NULL;
|
|
fseek(fp, 0, SEEK_END);
|
|
long len = ftell(fp);
|
|
fseek(fp, 0, SEEK_SET);
|
|
char *buf = (char*)malloc(len + 1);
|
|
size_t nread = fread(buf, 1, len, fp);
|
|
buf[nread] = '\0';
|
|
fclose(fp);
|
|
return buf;
|
|
}
|
|
|
|
static void rmdir_recursive(const char *path) {
|
|
char cmd[512];
|
|
snprintf(cmd, sizeof(cmd), "rm -rf %s", path);
|
|
system(cmd);
|
|
}
|
|
|
|
/* Case-insensitive substring search */
|
|
static int contains_ci(const char *haystack, const char *needle) {
|
|
if (!haystack || !needle) return 0;
|
|
int hlen = (int)strlen(haystack);
|
|
int nlen = (int)strlen(needle);
|
|
for (int i = 0; i <= hlen - nlen; i++) {
|
|
int match = 1;
|
|
for (int j = 0; j < nlen; j++) {
|
|
if (tolower((unsigned char)haystack[i + j]) !=
|
|
tolower((unsigned char)needle[j])) {
|
|
match = 0;
|
|
break;
|
|
}
|
|
}
|
|
if (match) return 1;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
/* Generate tokens greedily from a prefilled state. */
|
|
static char* generate(rx_model *model, rx_tokenizer *tok, rx_state *state,
|
|
tm_tensor *logits, int max_tokens, rx_config cfg) {
|
|
rx_sampling_params params = rx_sampling_defaults();
|
|
params.temperature = 0.0f;
|
|
|
|
uint32_t *gen_tokens = (uint32_t*)calloc(max_tokens, sizeof(uint32_t));
|
|
int n_gen = 0;
|
|
tm_tensor *cur_logits = logits;
|
|
|
|
for (int i = 0; i < max_tokens; i++) {
|
|
if (!cur_logits) break;
|
|
uint32_t token = rx_sample(cur_logits, ¶ms, NULL, cfg.vocab_size);
|
|
if (i > 0) tm_free(cur_logits);
|
|
if (token == 0) { cur_logits = NULL; break; }
|
|
gen_tokens[n_gen++] = token;
|
|
cur_logits = rx_forward(model, token, state);
|
|
}
|
|
if (cur_logits && cur_logits != logits) tm_free(cur_logits);
|
|
|
|
char *text = rx_tokenizer_decode(tok, gen_tokens, n_gen);
|
|
free(gen_tokens);
|
|
return text;
|
|
}
|
|
|
|
int main(int argc, char **argv) {
|
|
const char *model_path = NULL;
|
|
const char *vocab_path = NULL;
|
|
const char *filler_path = NULL;
|
|
int cache_size = 8;
|
|
int hm_top_k = 5;
|
|
int max_filler_chars = 2000;
|
|
int max_tokens = 48;
|
|
int quantize = 0;
|
|
|
|
for (int i = 1; i < argc; i++) {
|
|
if (strcmp(argv[i], "--model") == 0 && i + 1 < argc) model_path = argv[++i];
|
|
else if (strcmp(argv[i], "--vocab") == 0 && i + 1 < argc) vocab_path = argv[++i];
|
|
else if (strcmp(argv[i], "--filler") == 0 && i + 1 < argc) filler_path = argv[++i];
|
|
else if (strcmp(argv[i], "--cache-size") == 0 && i + 1 < argc) cache_size = atoi(argv[++i]);
|
|
else if (strcmp(argv[i], "--hm-top-k") == 0 && i + 1 < argc) hm_top_k = atoi(argv[++i]);
|
|
else if (strcmp(argv[i], "--max-filler") == 0 && i + 1 < argc) max_filler_chars = atoi(argv[++i]);
|
|
else if (strcmp(argv[i], "--max-tokens") == 0 && i + 1 < argc) max_tokens = atoi(argv[++i]);
|
|
else if (strcmp(argv[i], "--quantize") == 0) quantize = 1;
|
|
else if (strcmp(argv[i], "--help") == 0) {
|
|
fprintf(stderr, "Usage: %s --model PATH --vocab PATH --filler PATH [options]\n", argv[0]);
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
if (!model_path || !vocab_path || !filler_path) {
|
|
fprintf(stderr, "Error: --model, --vocab, and --filler are required\n");
|
|
return 1;
|
|
}
|
|
|
|
/* Load shared resources */
|
|
fprintf(stderr, "Loading tokenizer...\n");
|
|
rx_tokenizer *tok = rx_tokenizer_new(vocab_path);
|
|
if (!tok) { fprintf(stderr, "Failed to load tokenizer\n"); return 1; }
|
|
|
|
fprintf(stderr, "Loading model%s...\n", quantize ? " (INT8)" : "");
|
|
rx_model *model = quantize ? rx_model_load_q8(model_path)
|
|
: rx_model_load(model_path);
|
|
if (!model) { fprintf(stderr, "Failed to load model\n"); return 1; }
|
|
rx_config cfg = rx_model_config(model);
|
|
|
|
/* Load filler */
|
|
char *filler_raw = read_file(filler_path);
|
|
if (!filler_raw) { fprintf(stderr, "Failed to read filler\n"); return 1; }
|
|
if ((int)strlen(filler_raw) > max_filler_chars)
|
|
filler_raw[max_filler_chars] = '\0';
|
|
|
|
fprintf(stderr, "\n==========================================\n");
|
|
fprintf(stderr, "HDC Memory Viability — Test 5\n");
|
|
fprintf(stderr, "Multi-Fact Battery (%d facts)\n", N_FACTS);
|
|
fprintf(stderr, "==========================================\n\n");
|
|
|
|
int improved = 0, same = 0, worse = 0;
|
|
|
|
for (int f = 0; f < N_FACTS; f++) {
|
|
const fact_entry *fe = &FACTS[f];
|
|
fprintf(stderr, "--- Fact %d: \"%s\" ---\n", f + 1, fe->key_detail);
|
|
|
|
/* Build prompt */
|
|
char question_full[512];
|
|
snprintf(question_full, sizeof(question_full),
|
|
"\n\nQuestion: %s\nAnswer:", fe->question);
|
|
|
|
int pl = (int)(strlen(fe->fact) + 2 + strlen(filler_raw) + strlen(question_full) + 1);
|
|
char *prompt = (char*)malloc(pl);
|
|
snprintf(prompt, pl, "%s\n\n%s%s", fe->fact, filler_raw, question_full);
|
|
|
|
uint32_t *tokens = (uint32_t*)malloc(32768 * sizeof(uint32_t));
|
|
int n_tokens = rx_tokenizer_encode(tok, prompt, tokens, 32768);
|
|
|
|
/* Run A: no HM */
|
|
{
|
|
rx_hybridmem_config hm_cfg = rx_hybridmem_config_defaults();
|
|
hm_cfg.store_dir = NULL;
|
|
hm_cfg.max_cache_len = cache_size;
|
|
rx_state *state = rx_state_new_with_hybridmem(model, &hm_cfg);
|
|
|
|
tm_tensor *logits = NULL;
|
|
for (int i = 0; i < n_tokens; i++) {
|
|
if (logits) tm_free(logits);
|
|
logits = rx_forward(model, tokens[i], state);
|
|
}
|
|
char *out_a = generate(model, tok, state, logits, max_tokens, cfg);
|
|
tm_free(logits);
|
|
rx_state_free(state);
|
|
|
|
int hit_a = contains_ci(out_a, fe->key_detail);
|
|
fprintf(stderr, " A (no HM): %s [%s]\n",
|
|
out_a ? out_a : "(null)", hit_a ? "HIT" : "miss");
|
|
free(out_a);
|
|
|
|
/* Run B: with HM */
|
|
const char *hm_dir = "/tmp/hm_multi_fact";
|
|
rmdir_recursive(hm_dir);
|
|
|
|
hm_cfg.store_dir = hm_dir;
|
|
hm_cfg.max_cache_len = cache_size;
|
|
hm_cfg.top_k = hm_top_k;
|
|
rx_state *state_b = rx_state_new_with_hybridmem(model, &hm_cfg);
|
|
|
|
logits = NULL;
|
|
for (int i = 0; i < n_tokens; i++) {
|
|
if (logits) tm_free(logits);
|
|
logits = rx_forward(model, tokens[i], state_b);
|
|
}
|
|
char *out_b = generate(model, tok, state_b, logits, max_tokens, cfg);
|
|
tm_free(logits);
|
|
rx_state_free(state_b);
|
|
|
|
int hit_b = contains_ci(out_b, fe->key_detail);
|
|
fprintf(stderr, " B (HM): %s [%s]\n",
|
|
out_b ? out_b : "(null)", hit_b ? "HIT" : "miss");
|
|
free(out_b);
|
|
rmdir_recursive(hm_dir);
|
|
|
|
if (hit_b && !hit_a) improved++;
|
|
else if (!hit_b && hit_a) worse++;
|
|
else same++;
|
|
}
|
|
|
|
free(prompt);
|
|
free(tokens);
|
|
fprintf(stderr, "\n");
|
|
}
|
|
|
|
/* Summary */
|
|
fprintf(stderr, "==========================================\n");
|
|
fprintf(stderr, "SUMMARY\n");
|
|
fprintf(stderr, "==========================================\n");
|
|
fprintf(stderr, "Improved (B hit, A miss): %d/%d\n", improved, N_FACTS);
|
|
fprintf(stderr, "Same: %d/%d\n", same, N_FACTS);
|
|
fprintf(stderr, "Worse (A hit, B miss): %d/%d\n\n", worse, N_FACTS);
|
|
|
|
float gen_rate = (float)(improved + same) / N_FACTS * 100.0f;
|
|
fprintf(stderr, "VERDICT: ");
|
|
if (improved >= 6) {
|
|
fprintf(stderr, "PASS — %d/10 improved, generalization = %.0f%%\n", improved, gen_rate);
|
|
} else if (improved >= 3) {
|
|
fprintf(stderr, "MARGINAL — %d/10 improved, check which facts fail\n", improved);
|
|
} else {
|
|
fprintf(stderr, "FAIL — %d/10 improved, HybridMem doesn't generalize\n", improved);
|
|
}
|
|
if (worse > 0) {
|
|
fprintf(stderr, "WARNING: %d/10 WORSE with HybridMem — false retrieval may be harmful\n",
|
|
worse);
|
|
}
|
|
|
|
/* Cleanup */
|
|
free(filler_raw);
|
|
rx_model_free(model);
|
|
rx_tokenizer_free(tok);
|
|
|
|
fprintf(stderr, "\n==========================================\n");
|
|
return 0;
|
|
}
|