Fix todo: avoid relying on logits_all == true in perplexity_v2

common/common.cpp

@@ -2666,7 +2666,7 @@ void llama_batch_add(
     for (size_t i = 0; i < seq_ids.size(); ++i) {
         batch.seq_id[batch.n_tokens][i] = seq_ids[i];
     }
-    batch.logits  [batch.n_tokens] = logits;
+    batch.output  [batch.n_tokens] = logits;
 
     batch.n_tokens++;
 }

common/log.h

@@ -686,7 +686,7 @@ inline std::string LOG_BATCH_TOSTR_PRETTY(const C & ctx, const B & batch)
             << ":pos " << std::to_string(batch.pos[i])
             << ":n_seq_id  " << std::to_string(batch.n_seq_id[i])
             << ":seq_id " << std::to_string(batch.seq_id[i][0])
-            << ":logits " << std::to_string(batch.logits[i]);
+            << ":logits " << std::to_string(batch.output[i]);
     }
     buf << " ]";
 

examples/batched-bench/batched-bench.cpp

@@ -94,7 +94,7 @@ int main(int argc, char ** argv) {
                 batch.pos      + i,
                 batch.n_seq_id + i,
                 batch.seq_id   + i,
-                batch.logits   + i,
+                batch.output   + i,
                 0, 0, 0, // unused
             };
 
@@ -149,7 +149,7 @@ int main(int argc, char ** argv) {
                         llama_batch_add(batch, 0, i, { j }, false);
                     }
                 }
-                batch.logits[batch.n_tokens - 1] = true;
+                batch.output[batch.n_tokens - 1] = true;
 
                 const auto t_pp_start = ggml_time_us();
 

examples/batched.swift/Sources/main.swift

@@ -86,11 +86,11 @@ for (i, token) in tokens.enumerated() {
     if let seq_id = batch.seq_id[i] {
         seq_id[0] = 0
     }
-    batch.logits[i] = 0
+    batch.output[i] = 0
 }
 
 // llama_decode will output logits only for the last token of the prompt
-batch.logits[Int(batch.n_tokens) - 1] = 1
+batch.output[Int(batch.n_tokens) - 1] = 1
 
 if llama_decode(context, batch) != 0 {
     print("llama_decode() failed")
@@ -178,7 +178,7 @@ while n_cur <= n_len {
         if let seq_id = batch.seq_id[Int(batch.n_tokens)] {
             seq_id[0] = Int32(i)
         }
-        batch.logits[Int(batch.n_tokens)] = 1
+        batch.output[Int(batch.n_tokens)] = 1
 
         i_batch[i] = batch.n_tokens
 

examples/batched/batched.cpp

@@ -122,7 +122,7 @@ int main(int argc, char ** argv) {
     }
 
     // llama_decode will output logits only for the last token of the prompt
-    batch.logits[batch.n_tokens - 1] = true;
+    batch.output[batch.n_tokens - 1] = true;
 
     if (llama_decode(ctx, batch) != 0) {
         LOG_TEE("%s: llama_decode() failed\n", __func__);

examples/embedding/embedding.cpp

@@ -52,7 +52,7 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
     }
 
     for (int i = 0; i < batch.n_tokens; i++) {
-        if (!batch.logits[i]) {
+        if (!batch.output[i]) {
             continue;
         }
 

examples/gritlm/gritlm.cpp

@@ -102,21 +102,21 @@ static std::string generate(llama_context * ctx, const std::string & prompt, boo
     llama_set_embeddings(ctx, false);
     llama_set_causal_attn(ctx, true);
 
-    llama_batch bat = llama_batch_init(llama_n_batch(ctx), 0, 1);
+    llama_batch batch = llama_batch_init(llama_n_batch(ctx), 0, 1);
 
     std::vector<llama_token> inputs = llama_tokenize(mdl, prompt, false, true);
     int32_t i_current_token = 0;
 
     while (true) {
-        llama_batch_clear(bat);
+        llama_batch_clear(batch);
         auto n_inputs = (int32_t)inputs.size();
         for (int32_t i = 0; i < n_inputs; i++) {
-            llama_batch_add(bat, inputs[i], i_current_token++, { 0 }, i == n_inputs - 1);
+            llama_batch_add(batch, inputs[i], i_current_token++, { 0 }, i == n_inputs - 1);
         }
         inputs.clear();
 
-        llama_decode(ctx, bat);
-        auto logits = llama_get_logits_ith(ctx, bat.n_tokens - 1);
+        llama_decode(ctx, batch);
+        auto logits = llama_get_logits_ith(ctx, batch.n_tokens - 1);
 
         auto candidates = std::vector<llama_token_data>(llama_n_vocab(mdl));
         auto n_candidates = (int32_t)candidates.size();
@@ -145,7 +145,7 @@ static std::string generate(llama_context * ctx, const std::string & prompt, boo
         std::printf("\n");
     }
 
-    llama_batch_free(bat);
+    llama_batch_free(batch);
 
     return result;
 }

examples/imatrix/imatrix.cpp

@@ -513,7 +513,7 @@ static bool compute_imatrix(llama_context * ctx, const gpt_params & params) {
                 tokens[batch_start] = llama_token_bos(llama_get_model(ctx));
             }
 
-            // TODO: use batch.logits to save computations instead of relying on logits_all == true
+            // TODO: use batch.output to save computations instead of relying on logits_all == true
             if (llama_decode(ctx, llama_batch_get_one(tokens.data() + batch_start, batch_size, j * n_batch, 0))) {
                 fprintf(stderr, "%s : failed to eval\n", __func__);
                 return false;

examples/llama.android/llama/src/main/cpp/llama-android.cpp

@@ -193,7 +193,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
             llama_batch_add(*batch, 0, i, { 0 }, false);
         }
 
-        batch->logits[batch->n_tokens - 1] = true;
+        batch->output[batch->n_tokens - 1] = true;
         llama_kv_cache_clear(context);
 
         const auto t_pp_start = ggml_time_us();
@@ -306,7 +306,7 @@ Java_android_llama_cpp_LLamaAndroid_new_1batch(JNIEnv *, jobject, jint n_tokens,
     for (int i = 0; i < n_tokens; ++i) {
         batch->seq_id[i] = (llama_seq_id *) malloc(sizeof(llama_seq_id) * n_seq_max);
     }
-    batch->logits   = (int8_t *)        malloc(sizeof(int8_t)         * n_tokens);
+    batch->output   = (int8_t *)        malloc(sizeof(int8_t)         * n_tokens);
 
     return reinterpret_cast<jlong>(batch);
 }
@@ -363,7 +363,7 @@ Java_android_llama_cpp_LLamaAndroid_completion_1init(
     }
 
     // llama_decode will output logits only for the last token of the prompt
-    batch->logits[batch->n_tokens - 1] = true;
+    batch->output[batch->n_tokens - 1] = true;
 
     if (llama_decode(context, *batch) != 0) {
         LOGe("llama_decode() failed");

examples/llama.swiftui/llama.cpp.swift/LibLlama.swift

@@ -16,7 +16,7 @@ func llama_batch_add(_ batch: inout llama_batch, _ id: llama_token, _ pos: llama
     for i in 0..<seq_ids.count {
         batch.seq_id[Int(batch.n_tokens)]![Int(i)] = seq_ids[i]
     }
-    batch.logits  [Int(batch.n_tokens)] = logits ? 1 : 0
+    batch.output  [Int(batch.n_tokens)] = logits ? 1 : 0
 
     batch.n_tokens += 1
 }
@@ -132,7 +132,7 @@ actor LlamaContext {
             let i = Int(i1)
             llama_batch_add(&batch, tokens_list[i], Int32(i), [0], false)
         }
-        batch.logits[Int(batch.n_tokens) - 1] = 1 // true
+        batch.output[Int(batch.n_tokens) - 1] = 1 // true
 
         if llama_decode(context, batch) != 0 {
             print("llama_decode() failed")
@@ -214,7 +214,7 @@ actor LlamaContext {
             for i in 0..<n_tokens {
                 llama_batch_add(&batch, 0, Int32(i), [0], false)
             }
-            batch.logits[Int(batch.n_tokens) - 1] = 1 // true
+            batch.output[Int(batch.n_tokens) - 1] = 1 // true
 
             llama_kv_cache_clear(context)
 

examples/parallel/parallel.cpp

@@ -265,7 +265,7 @@ int main(int argc, char ** argv) {
 
                     // extract the logits only for the last token
                     if (batch.n_tokens > 0) {
-                        batch.logits[batch.n_tokens - 1] = true;
+                        batch.output[batch.n_tokens - 1] = true;
                     }
 
                     client.n_prompt  = tokens_prompt.size();
@@ -308,7 +308,7 @@ int main(int argc, char ** argv) {
                 batch.pos      + i,
                 batch.n_seq_id + i,
                 batch.seq_id   + i,
-                batch.logits   + i,
+                batch.output   + i,
                 0, 0, 0, // unused
             };
 

examples/passkey/passkey.cpp

@@ -140,7 +140,7 @@ int main(int argc, char ** argv) {
         }
 
         if (i + n_batch >= n_tokens_all) {
-            batch.logits[batch.n_tokens - 1] = true;
+            batch.output[batch.n_tokens - 1] = true;
         }
 
         if (llama_decode(ctx, batch) != 0) {
@@ -174,7 +174,7 @@ int main(int argc, char ** argv) {
         }
 
         if (i + n_batch >= n_tokens_all) {
-            batch.logits[batch.n_tokens - 1] = true;
+            batch.output[batch.n_tokens - 1] = true;
         }
 
         if (llama_decode(ctx, batch) != 0) {

examples/perplexity/perplexity.cpp

@@ -367,17 +367,15 @@ static results_perplexity perplexity_v2(llama_context * ctx, const gpt_params &
         return {tokens, -1, logit_history, prob_history};
     }
 
-    const int calc_chunk = n_ctx;
+    fprintf(stderr, "%s: have %zu tokens. Calculation chunk = %d\n", __func__, tokens.size(), n_ctx);
 
-    fprintf(stderr, "%s: have %zu tokens. Calculation chunk = %d\n", __func__, tokens.size(), calc_chunk);
-
-    if (int(tokens.size()) <= calc_chunk) {
+    if (int(tokens.size()) <= n_ctx) {
         fprintf(stderr, "%s: there are only %zu tokens, this is not enough for a context size of %d and stride %d\n",__func__,
                 tokens.size(), n_ctx, params.ppl_stride);
         return {tokens, -1, logit_history, prob_history};
     }
 
-    const int n_chunk_max = (tokens.size() - calc_chunk + params.ppl_stride - 1)  / params.ppl_stride;
+    const int n_chunk_max = (tokens.size() - n_ctx + params.ppl_stride - 1)  / params.ppl_stride;
 
     const int n_chunk = params.n_chunks < 0 ? n_chunk_max : std::min(params.n_chunks, n_chunk_max);
     const int n_vocab = llama_n_vocab(llama_get_model(ctx));
@@ -386,13 +384,13 @@ static results_perplexity perplexity_v2(llama_context * ctx, const gpt_params &
     int count = 0;
     double nll = 0.0;
 
+    const int num_batches = (n_ctx + n_batch - 1) / n_batch;
+
     fprintf(stderr, "%s: calculating perplexity over %d chunks, batch_size=%d\n", __func__, n_chunk, n_batch);
 
     for (int i = 0; i < n_chunk; ++i) {
         const int start =     i * params.ppl_stride;
-        const int end   = start + calc_chunk;
-
-        const int num_batches = (calc_chunk + n_batch - 1) / n_batch;
+        const int end   = start + n_ctx;
         //fprintf(stderr, "%s: evaluating %d...%d using %d batches\n", __func__, start, end, num_batches);
 
         std::vector<float> logits;
@@ -406,13 +404,27 @@ static results_perplexity perplexity_v2(llama_context * ctx, const gpt_params &
             const int batch_start = start + j * n_batch;
             const int batch_size  = std::min(end - batch_start, n_batch);
 
+            llama_batch batch = llama_batch_init(batch_size, 0, 1);
+            for (int k = 0; k < batch_size; ++k) {
+                const int idx = batch_start + k;
+                batch.token  [k] = tokens[idx];
+                batch.output [k] = 1;
+            }
+            batch.n_tokens   = batch_size;
+            batch.pos        = nullptr;
+            batch.n_seq_id   = nullptr;
+            batch.seq_id     = nullptr;
+            batch.all_pos_0  = j*n_batch;
+            batch.all_pos_1  = 1;
+            batch.all_seq_id = 0;
+
             //fprintf(stderr, "    Batch %d: starts at %d, size is %d, n_past is %d\n",j,batch_start,batch_size,j * n_batch);
-            // TODO: use llama_batch.logits instead of relying on logits_all == true
-            if (llama_decode(ctx, llama_batch_get_one(tokens.data() + batch_start, batch_size, j * n_batch, 0))) {
+            if (llama_decode(ctx, batch)) {
                 //fprintf(stderr, "%s : failed to eval\n", __func__);
                 return {tokens, -1, logit_history, prob_history};
             }
 
+            llama_batch_free(batch);
             // save original token and restore it after eval
             const auto token_org = tokens[batch_start];
 
@@ -601,9 +613,9 @@ static results_perplexity perplexity(llama_context * ctx, const gpt_params & par
                     batch.pos     [idx]    = j*n_batch + k;
                     batch.n_seq_id[idx]    = 1;
                     batch.seq_id  [idx][0] = seq;
-                    batch.logits  [idx]    = batch.pos[idx] >= first ? 1 : 0;
+                    batch.output  [idx]    = batch.pos[idx] >= first ? 1 : 0;
 
-                    n_outputs += batch.logits[idx] != 0;
+                    n_outputs += batch.output[idx] != 0;
                 }
                 batch.n_tokens += batch_size;
 
@@ -697,7 +709,7 @@ static bool decode_helper(llama_context * ctx, llama_batch & batch, std::vector<
             batch.pos      + i,
             batch.n_seq_id + i,
             batch.seq_id   + i,
-            batch.logits   + i,
+            batch.output   + i,
             0, 0, 0, // unused
         };
 
@@ -709,7 +721,7 @@ static bool decode_helper(llama_context * ctx, llama_batch & batch, std::vector<
 
         int n_outputs = 0;
         for (int i = 0; i < n_tokens; ++i) {
-            n_outputs += batch_view.logits[i] != 0;
+            n_outputs += batch_view.output[i] != 0;
         }
 
         memcpy(batch_logits.data() + prev_outputs*n_vocab, llama_get_logits(ctx), n_outputs*n_vocab*sizeof(float));
@@ -917,7 +929,7 @@ static void hellaswag_score(llama_context * ctx, const gpt_params & params) {
             for (size_t i = 0; i < hs_cur.common_prefix; ++i) {
                 llama_batch_add(batch, hs_cur.seq_tokens[0][i], i, { s0 + 0, s0 + 1, s0 + 2, s0 + 3 }, false);
             }
-            batch.logits[batch.n_tokens - 1] = true; // we need logits for the last token of the common prefix
+            batch.output[batch.n_tokens - 1] = true; // we need logits for the last token of the common prefix
             n_logits += 1;
 
             for (int s = 0; s < 4; ++s) {
@@ -1196,7 +1208,7 @@ static void winogrande_score(llama_context * ctx, const gpt_params & params) {
             for (size_t i = 0; i < data[i1].common_prefix; ++i) {
                 llama_batch_add(batch, data[i1].seq_tokens[0][i], i, { s0 + 0, s0 + 1 }, false);
             }
-            batch.logits[batch.n_tokens - 1] = true;
+            batch.output[batch.n_tokens - 1] = true;
             n_logits += 1;
 
             for (int s = 0; s < 2; ++s) {
@@ -1565,7 +1577,7 @@ static void multiple_choice_score(llama_context * ctx, const gpt_params & params
                 //llama_batch_add(batch, cur_task.seq_tokens[0][i], i, { s0 + 0, s0 + 1, s0 + 2, s0 + 3}, false);
                 llama_batch_add(batch, cur_task.seq_tokens[0][i], i, batch_indeces, false);
             }
-            batch.logits[batch.n_tokens - 1] = true; // we need logits for the last token of the common prefix
+            batch.output[batch.n_tokens - 1] = true; // we need logits for the last token of the common prefix
             n_logits += 1;
 
             for (int s = 0; s < int(cur_task.seq_tokens.size()); ++s) {
@@ -1794,7 +1806,7 @@ static void kl_divergence(llama_context * ctx, const gpt_params & params) {
                 tokens[batch_start] = llama_token_bos(llama_get_model(ctx));
             }
 
-            // TODO: use llama_batch.logits instead of relying on logits_all == true
+            // TODO: use llama_batch.output instead of relying on logits_all == true
             if (llama_decode(ctx, llama_batch_get_one(tokens.data() + batch_start, batch_size, j * n_batch, 0))) {
                 fprintf(stderr, "%s : failed to eval\n", __func__);
                 return;

examples/retrieval/retrieval.cpp

@@ -91,7 +91,7 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
     }
 
     for (int i = 0; i < batch.n_tokens; i++) {
-        if (!batch.logits[i]) {
+        if (!batch.output[i]) {
             continue;
         }
 

examples/server/server.cpp

@@ -1480,7 +1480,7 @@ struct server_context {
         std::vector<float> embd_res(n_embd, 0.0f);
 
         for (int i = 0; i < batch.n_tokens; ++i) {
-            if (!batch.logits[i] || batch.seq_id[i][0] != slot.id + 1) {
+            if (!batch.output[i] || batch.seq_id[i][0] != slot.id + 1) {
                 continue;
             }
 
@@ -2269,7 +2269,7 @@ struct server_context {
                         GGML_ASSERT(batch.n_tokens > 0);
 
                         // extract the logits only for the last token
-                        batch.logits[batch.n_tokens - 1] = true;
+                        batch.output[batch.n_tokens - 1] = true;
 
                         slot.n_decoded = 0;
                         slot.i_batch   = batch.n_tokens - 1;
@@ -2341,7 +2341,7 @@ struct server_context {
                 batch.pos      + i,
                 batch.n_seq_id + i,
                 batch.seq_id   + i,
-                batch.logits   + i,
+                batch.output   + i,
                 0, 0, 0, // unused
             };
 

examples/simple/simple.cpp

@@ -93,7 +93,7 @@ int main(int argc, char ** argv) {
     }
 
     // llama_decode will output logits only for the last token of the prompt
-    batch.logits[batch.n_tokens - 1] = true;
+    batch.output[batch.n_tokens - 1] = true;
 
     if (llama_decode(ctx, batch) != 0) {
         LOG_TEE("%s: llama_decode() failed\n", __func__);