mrq/vall-e
5
9
Fork 0
Code Issues 8 Pull Requests Packages Projects Releases Wiki Activity

ugh

Browse Source
This commit is contained in:
mrq 2025-04-05 18:58:25 -05:00
parent b6692ce3de
commit c34763769a
1 changed files with 20 additions and 40 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
vall_e.cpp/vall_e.cpp
View File

@ -89,28 +89,7 @@ std::vector<float> read_2d_tensor( struct ggml_tensor* tensor ) {
return res;
}
/*
ggml_tensor* view_2d_tensor( struct ggml_tensor* tensor, int32_t start, int32_t end, int32_t dim ) {
// to-do: implement other dim
if ( start < 0 ) start = tensor->ne[1] + start;
if ( end < 0 ) end = tensor->ne[1] + end;
ggml_tensor* res = new ggml_tensor();
memcpy( res, tensor, sizeof(ggml_tensor) );
res->op = GGML_OP_VIEW;
res->src[0] = tensor;
res->data += res->nb[1] * start;
res->ne[1] = end - start;
for (int i = 2; i < GGML_MAX_DIMS; i++) {
res->nb[i] = res->nb[i - 1] * res->ne[i - 1];
}
return res;
}
*/
ggml_tensor* view_2d_tensor( struct ggml_context* ctx, struct ggml_tensor* tensor, int32_t start, int32_t end, int32_t dim ) {
// to-do: implement other dim
if ( start < 0 ) start = tensor->ne[1] + start;
@ -601,13 +580,14 @@ std::vector<token_t> generate( vall_e_context_t* ctx, vall_e_inputs_t& inputs, i
int32_t seq_len = n_outputs;
int32_t top_k = 0;
float top_p = 1.0;
float temperature = 1.0f;
float cfg_strength = 2.5f;
float temperature = 1.5f;
float cfg_strength = 3.0f;
float start_noise = 0.0f;
float end_noise = 1.0f;
bool annealed_sampling = true;
bool remasking = true;
float cfg_rescale = 0.75f;
bool entropy_scoring = true;
// fill with masked tokens
output_tokens.clear();
@ -621,7 +601,7 @@ std::vector<token_t> generate( vall_e_context_t* ctx, vall_e_inputs_t& inputs, i
llama_batch null_batch = llama_batch_init( ctx->params.ctx_size, ctx->io_map->n_embd, ctx->params.ctx_size );
// token scores to reference for masking
std::vector<float> scores(n_outputs, 1.0);
std::vector<float> scores(n_outputs, entropy_scoring ? 0.0 : 1.0);
// do one step on many tokens
for ( auto step = 0; step < steps; ++step ) {
@ -635,7 +615,7 @@ std::vector<token_t> generate( vall_e_context_t* ctx, vall_e_inputs_t& inputs, i
float sampling_cfg_strength = annealed_sampling ? timestep * cfg_strength : cfg_strength;
float noise_p = cos( timestep * PI * 0.5f );
float remask_p = remasking ? 0.5f / steps : 0.0f;
float remask_p = remasking ? 1.0f / (steps * 2.0f) : 0.0f;
int32_t n_masked_tokens = (noise_p + remask_p) * seq_len;
if ( n_masked_tokens < 1 ) {
@ -651,7 +631,9 @@ std::vector<token_t> generate( vall_e_context_t* ctx, vall_e_inputs_t& inputs, i
std::vector<score_t> sorted_scores( n_outputs );
for ( auto i = 0; i < n_outputs; ++i ) sorted_scores[i] = { i, scores[i] };
std::sort(sorted_scores.begin(), sorted_scores.end());
// std::reverse(sorted_scores.begin(), sorted_scores.end());
if ( entropy_scoring) {
std::reverse(sorted_scores.begin(), sorted_scores.end());
}
// and top-k pick the worst scores
for ( auto i = 0; i < n_masked_tokens; ++i ) {
@ -669,8 +651,8 @@ std::vector<token_t> generate( vall_e_context_t* ctx, vall_e_inputs_t& inputs, i
inputs.resp[0] = output_tokens;
fill_batch( batch, inputs, *ctx->io_map, mode );
// update null batch
null_input.resp[0] = output_tokens;
null_batch.n_tokens = 0;
null_input.resp[0] = output_tokens;
fill_batch( null_batch, inputs, *ctx->io_map, mode );
// cfg decode
@ -707,7 +689,7 @@ std::vector<token_t> generate( vall_e_context_t* ctx, vall_e_inputs_t& inputs, i
for ( auto idx = 0; idx < n_outputs; ++idx ) {
// skip if not masked
if ( !is_masked[idx] ) {
scores[idx] = 0.0;
scores[idx] = entropy_scoring ? 0.0 : 1.0;
continue;
}
@ -716,6 +698,7 @@ std::vector<token_t> generate( vall_e_context_t* ctx, vall_e_inputs_t& inputs, i
// perform softmax before modifying logits
std::vector<float> softmaxed = soft_max( n_vocab, logit );
int32_t t_u = std::distance( softmaxed.begin(), std::max_element(softmaxed.begin(), softmaxed.end()) );
std::vector<float> summed(n_vocab);
for (int i = 0; i < n_vocab; i++) {
@ -739,19 +722,16 @@ std::vector<token_t> generate( vall_e_context_t* ctx, vall_e_inputs_t& inputs, i
// store token if it was masked
output_tokens[idx] = t;
// update score if it was masked
// this is actually wrong
scores[idx] = 1.0 - softmaxed[t]; // invert so we pick the worst tokens later
// this seems to work better
/*
float entropy = 0.f;
for (int v = 0; v < n_vocab; ++v ) {
float p = softmaxed[v];
if (p > 0) entropy -= p * std::log(p + 1e-9);
if ( entropy_scoring ) {
float entropy = 0.f;
for (int v = 0; v < n_vocab; ++v ) {
float p = softmaxed[v];
if (p > 0) entropy -= p * std::log(p + 1e-9);
}
scores[idx] = entropy / std::log(n_vocab); // normalize [0–1]
} else {
scores[idx] = softmaxed[t_u]; // invert so we pick the worst tokens later
}
scores[idx] = entropy / std::log(n_vocab); // normalize [0–1]
*/
}
llama_sampler_free(smpl);