llama : greatly reduce output buffer memory usage (llama/6122)

* llama : greatly reduce logits memory usage

* llama : more compact state saving and reloading

* llama : fix lctx.n_outputs not being set before building graph

* perplexity : adapt to the logits API changes

* perplexity : fix Winogrande, use correct logits for second choice start

The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.

The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.

This is simpler now, and the outlier scores aren't there anymore.

* perplexity : normalize spaces and punctuation in Winogrande sentences

* llama : fix embedding conditions

* llama : fix llama_get_embeddings_ith when the resulting id is 0

* llama : fix wrong n_outputs in llama_set_inputs

A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.

* llama : when saving the state, recalculate n_outputs

This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.

* llama : fix not-skipping outputs of non-causal models

* llama : fix running a batch with n_outputs == 0

It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.

* llama : keep same graph topology even when n_outputs == 0

* ggml : saner ggml_can_repeat with empty tensors

*  ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1

* ggml : do not multi-thread ops returning empty tensors

* ggml : make ggml_is_empty public and work with views

* llama : use a vector for ctx->output_ids

* llama : rework reallocation logic for llama_output_reserve

Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.

* ggml : skip empty tensors in all backends

* llama : fix llama_output_reserve nullptr deref when new_size is 0

* perplexity : make Winogrande work as it does on master

The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.

* llama : clearer error messages for invalid logits or embeddings ids

* llama : assert all models that can have inp_out_ids

Since the graph topology is now constant, this presence check
can be done even when there are no outputs.

* llama : assert logits and embd buffers exist before writing to them

* llama : handle errors from llama_output_reserve at call sites

* perplexity : make hellaswag and multiple-choice outputs identical to master

Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.

This will probably be changed back in the future to make these benchmarks
a tiny bit faster.

* perplexity : fix division by zero when using less than 100 multiple-choice tasks

* llama : allow loading state saved with a different ctx size

When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.

Doing this enables the use-case of extending or shrinking the context size
of a saved session.

This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.

* llama : minor

ggml-ci

* readme : update recent API changes, and warn about Vulkan

---------

Co-authored-by: Georgi Gerganov <redacted>

diff --git a/include/ggml/ggml.h b/include/ggml/ggml.h
index 425c9b6ab2d6df4924ade1895be332dacfe50ca0..5d4a4ceb65c7e106bf2008ba82089fe8d4d7a83f 100644 (file)
--- a/include/ggml/ggml.h
+++ b/include/ggml/ggml.h
@@ -750,6 +750,7 @@ extern "C" {
     GGML_API GGML_CALL bool ggml_is_transposed(const struct ggml_tensor * tensor);
     GGML_API GGML_CALL bool ggml_is_contiguous(const struct ggml_tensor * tensor);
     GGML_API GGML_CALL bool ggml_is_permuted  (const struct ggml_tensor * tensor);
+    GGML_API GGML_CALL bool ggml_is_empty     (const struct ggml_tensor * tensor);
     GGML_API           bool ggml_is_scalar    (const struct ggml_tensor * tensor);
     GGML_API           bool ggml_is_vector    (const struct ggml_tensor * tensor);
     GGML_API           bool ggml_is_matrix    (const struct ggml_tensor * tensor);

diff --git a/src/ggml-cuda.cu b/src/ggml-cuda.cu
index 48232b6e18d6ce98f6c57dabdafc44297df9ff7b..be8e33a56c40f45dac8807b419925379c72f3c0f 100644 (file)
--- a/src/ggml-cuda.cu
+++ b/src/ggml-cuda.cu
@@ -2505,7 +2505,7 @@ GGML_CALL static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t
     for (int i = 0; i < cgraph->n_nodes; i++) {
         ggml_tensor * node = cgraph->nodes[i];
 
-        if (node->op == GGML_OP_RESHAPE || node->op == GGML_OP_TRANSPOSE || node->op == GGML_OP_VIEW || node->op == GGML_OP_PERMUTE || node->op == GGML_OP_NONE) {
+        if (ggml_is_empty(node) || node->op == GGML_OP_RESHAPE || node->op == GGML_OP_TRANSPOSE || node->op == GGML_OP_VIEW || node->op == GGML_OP_PERMUTE || node->op == GGML_OP_NONE) {
             continue;
         }
 

diff --git a/src/ggml-kompute.cpp b/src/ggml-kompute.cpp
index 81dd5067864ce73033a003efdcf54cb44eab4eea..407062e6fd47625d6cb2f78e17649f64387d2866 100644 (file)
--- a/src/ggml-kompute.cpp
+++ b/src/ggml-kompute.cpp
@@ -1430,6 +1430,10 @@ static void ggml_vk_graph_compute(struct ggml_kompute_context * ctx, struct ggml
             struct ggml_tensor * dst = gf->nodes[i];
             GGML_ASSERT(dst->data != nullptr);
 
+            if (ggml_is_empty(dst)) {
+                continue;
+            }
+
             switch (dst->op) {
                 case GGML_OP_NONE:
                 case GGML_OP_RESHAPE:

diff --git a/src/ggml-metal.m b/src/ggml-metal.m
index cbe22aa3792b460d376a69f26fa8ecd8f1a4f007..a08abbc2918028cc178d997a67946b8030295fd6 100644 (file)
--- a/src/ggml-metal.m
+++ b/src/ggml-metal.m
@@ -847,6 +847,10 @@ static enum ggml_status ggml_metal_graph_compute(
             struct ggml_tensor * src2 = gf->nodes[i]->src[2];
             struct ggml_tensor * dst  = gf->nodes[i];
 
+            if (ggml_is_empty(dst)) {
+                continue;
+            }
+
             switch (dst->op) {
                 case GGML_OP_NONE:
                 case GGML_OP_RESHAPE:

diff --git a/src/ggml-opencl.cpp b/src/ggml-opencl.cpp
index aa73d67df84b04197700e727e2b02693edf84591..b3f8b7eaf0a3b8f7e7ebfa5389611e9b1a165cf8 100644 (file)
--- a/src/ggml-opencl.cpp
+++ b/src/ggml-opencl.cpp
@@ -2234,6 +2234,11 @@ static ggml_backend_buffer_type_t ggml_backend_opencl_get_default_buffer_type(gg
 static ggml_status ggml_backend_opencl_graph_compute(ggml_backend_t backend, ggml_cgraph * graph) {
     for (int i = 0; i < graph->n_nodes; ++i) {
         ggml_tensor * node = graph->nodes[i];
+
+        if (ggml_is_empty(node)) {
+            continue;
+        }
+
         switch (node->op) {
             case GGML_OP_MUL_MAT:
                 ggml_cl_mul_mat(node->src[0], node->src[1], node, nullptr, 0);

diff --git a/src/ggml-sycl.cpp b/src/ggml-sycl.cpp
index fc4d2964ccac9621e28d142cbfea8f2aa0f15c74..789ba97bfba39ac8a7a9b0064bdf549da1177f76 100644 (file)
--- a/src/ggml-sycl.cpp
+++ b/src/ggml-sycl.cpp
@@ -16973,7 +16973,7 @@ GGML_CALL static ggml_status ggml_backend_sycl_graph_compute(ggml_backend_t back
     params.ith = 0;
     for (int i = 0; i < cgraph->n_nodes; i++) {
         ggml_tensor * node = cgraph->nodes[i];
-        if (node->op == GGML_OP_RESHAPE || node->op == GGML_OP_TRANSPOSE || node->op == GGML_OP_VIEW || node->op == GGML_OP_PERMUTE || node->op == GGML_OP_NONE) {
+        if (ggml_is_empty(node) || node->op == GGML_OP_RESHAPE || node->op == GGML_OP_TRANSPOSE || node->op == GGML_OP_VIEW || node->op == GGML_OP_PERMUTE || node->op == GGML_OP_NONE) {
             continue;
         }
 #ifndef NDEBUG

diff --git a/src/ggml-vulkan.cpp b/src/ggml-vulkan.cpp
index cbceaa19fbacde29d6b950c504edeccf92c11989..521a1314b35655dc6fc1e3ef42e013137eae1347 100644 (file)
--- a/src/ggml-vulkan.cpp
+++ b/src/ggml-vulkan.cpp
@@ -5566,7 +5566,7 @@ GGML_CALL static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backen
     for (int i = 0; i < cgraph->n_nodes; i++) {
         ggml_tensor * node = cgraph->nodes[i];
 
-        if (node->op == GGML_OP_RESHAPE || node->op == GGML_OP_TRANSPOSE || node->op == GGML_OP_VIEW || node->op == GGML_OP_PERMUTE || node->op == GGML_OP_NONE) {
+        if (ggml_is_empty(node) || node->op == GGML_OP_RESHAPE || node->op == GGML_OP_TRANSPOSE || node->op == GGML_OP_VIEW || node->op == GGML_OP_PERMUTE || node->op == GGML_OP_NONE) {
             continue;
         }
 

diff --git a/src/ggml.c b/src/ggml.c
index a86b41c158558820c5921e0ad2c827dea874016e..eb469d0f7953d514bdf14a99009fdda0c2b00453 100644 (file)
--- a/src/ggml.c
+++ b/src/ggml.c
@@ -2607,6 +2607,16 @@ static inline bool ggml_is_padded_1d(const struct ggml_tensor * tensor) {
         tensor->nb[3] == tensor->nb[2]*tensor->ne[2];
 }
 
+GGML_CALL bool ggml_is_empty(const struct ggml_tensor * tensor) {
+    for (int i = 0; i < GGML_MAX_DIMS; ++i) {
+        if (tensor->ne[i] == 0) {
+            // empty if any dimension has no elements
+            return true;
+        }
+    }
+    return false;
+}
+
 bool ggml_are_same_shape(const struct ggml_tensor * t0, const struct ggml_tensor * t1) {
     static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
 
@@ -2621,7 +2631,7 @@ bool ggml_are_same_shape(const struct ggml_tensor * t0, const struct ggml_tensor
 static inline bool ggml_can_repeat(const struct ggml_tensor * t0, const struct ggml_tensor * t1) {
     static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
 
-    return
+    return ggml_is_empty(t0) ? ggml_is_empty(t1) :
         (t1->ne[0]%t0->ne[0] == 0) &&
         (t1->ne[1]%t0->ne[1] == 0) &&
         (t1->ne[2]%t0->ne[2] == 0) &&
@@ -16114,7 +16124,7 @@ static void ggml_compute_forward_cross_entropy_loss_back(
 static void ggml_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor) {
     GGML_ASSERT(params);
 
-    if (tensor->op == GGML_OP_NONE) {
+    if (tensor->op == GGML_OP_NONE || ggml_is_empty(tensor)) {
         return;
     }
 
@@ -17983,6 +17993,12 @@ static void ggml_graph_compute_perf_stats_node(struct ggml_tensor * node, const
 static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads, int n_cur_threads) {
     int n_tasks = 0;
 
+    if (ggml_is_empty(node)) {
+        // no need to multi-thread a no-op
+        n_tasks = 1;
+        return n_tasks;
+    }
+
     switch (node->op) {
         case GGML_OP_CPY:
         case GGML_OP_DUP: