mirror of
https://github.com/ollama/ollama.git
synced 2026-04-24 17:55:43 +02:00
update vendored llama.cpp and ggml (#11823)
* TEMPORARY: Update the llama.cpp upstream to my fork's Granite Four branch
This will be redone once my branch is merged upstream in llama.cpp
* feat: Update all patches
There are a number that are no longer needed at all:
- 0003-embeddings: Embeddings entirely overhauled on master
- 0008-ensure-KV-cache-is-fully-defragmented: KV caching entirely
overhauled on master
- 0019-metal-add-mean-kernel-14267: Merged upstream
- 0020-CUDA-add-mean-operation-14313: Merged upstream
* feat: Sync llama.cpp and ggml
* fix: Update rsync-filter for all moved/new/removed files
* fix: Add files missing from sync
* fix: Update ggml rsync-filter for new ggml-cpu/arch subdirs
* fix: Add ggml files missing from sync
* fix: Narrow llama.cpp rsync-filter to not include mtmd main tool cpp files
* fix: Remove mtmd main cpp files
* fix: Add missing include in sampling_ext.cpp
* fix: Update llama.go to use mtmd instead of clip/llava
* fix: Add patch for mtmd_input_text
* chore: Ignore *.patched in the patch directory
* fix: Fix support for arch-specific ggml-cpu source files with new arrangement
In https://github.com/ggml-org/llama.cpp/pull/13892, all arch-specific
implementations were split out into a nested tree structure under
ggml-cpu/arch. This conflicts with standard CGO layout where all
arch-specific source files are expected to live in the same directory as
the parent go module and use suffixes based on GOOS and GOARCH. As such,
there were really two options for getting this to work:
1. Add a patch on top of the GGML sync to rearrange the files to match the
GO layout convention
2. Use CGO directives to conditionally include the nested source files in
the compilation units
This commit does (2) in order to minimize the set of changes needed on top
of the upstream file layout. To get this to work, there are two key things
needed:
1. In cpu.go, #cgo directives are added to explicitly set __${GOARCH}__ in
the preprocessor directives
2. In arch-impls.c|cpp, use an #ifdef | #elif defined | #endif chain to
explicitly include the .c|.cpp files for the given architecture from the
nested directory
* fix: Use mtmd_helper to correctly load the bitmap for the image
* fix: Apply patch for mtmd_text_input
* fix: Add missing stb to llama.cpp rsync-filter
* fix: Add sync'ed stb vendored header
* fix: Use c++17 and include vendor for go wrapper modules
* fix: Update patch 0015 for upstream implementation of uuid
* feat: Bump to the latest tip of the branch
* fix: Update patches for bump
* feat: Bump back to the cenral repo and point at the latest master
This includes granite 4 and a number of other model architectures!
* fix: Revert changes to ggml export GPU UUID patch
* fix: Add patch for GGML_VERSION and GGML_COMMIT constants
* feat: Sync all patched code
* build: Include cmake/common.cmake in ggml sync
* build: Add top-level include for GNUINstallDirs in CMakeLists.txt
This is used to populate CMAKE_INSTALL_BINDIR
* fix: Add a patch to avoid power throttling API on non-msvc windows builds
* fix: Sync patch changes for ggml-cpu.c
* feat: Bump llama.cpp to 4a4f42
This picks up support for Kimi K2 and PLaMO-2
* feat: Sync llama.cpp
* fix: Handle multi-chunk image encodings from mtmd
* fix: Re-number patches after merge with `main`
* feat: Bump to 41e78c in the makefile
* fix: Fix Solar and argsort/copy patches after bump
* fix: Remove Gemma3n CUDA Graphs patch
It was implemented upstream:
https://github.com/ggml-org/llama.cpp/pull/14741
* feat: Sync llama.cpp / ggml after latest bump
* build: Remove unnecessary CFLAGS definitions in cpu.go
* fix: Remove unnecessary additions in the rsync-filter
* fix: Remove unused vendored code for chat template parsing
* Revert "fix: Remove Gemma3n CUDA Graphs patch"
This reverts commit d724caced3.
* fix: Update 0020 CUDA Graphs for gemma3n to keep both llama.cpp and ollama fixes
https://github.com/ollama/ollama/pull/11195#issuecomment-3137312394
* fix: Sync ggml-cuda.cu after keeping both style cuda graph fixes for gemma3n
* unwind mxfp4 patch
Prepare to bump ggml with their impl for mxfp4
* bump
* fix windows build error
* Convert tensors at load time
Repack the mxfp4 tensors as ggmls kernels expect them to be.
* convert mlp bf16 to f32
* buffer the conversion better
* reshape earlier
* openai swiglu
* add ids
* split qkv, gate_up
* fix nested alt tags
* fast attention
* remove debug messages
* fix lint
* remove redundant test
* remap values only if source/target are different
* add back i32->i32 copy
* refactor cpu quants
* clean up vendor
* update patch instructions
* clean up patches
* remove webgpu
* update mem
* also handle gpt-oss
* revert convert changes
---------
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Daniel Hiltgen <daniel@ollama.com>
This commit is contained in:
Michael Yang
GitHub
413
llama/llama.cpp/src/llama-kv-cache.h
vendored
413
llama/llama.cpp/src/llama-kv-cache.h
vendored
@@ -1,413 +0,0 @@
|
||||
#pragma once
|
||||
|
||||
#include "llama.h"
|
||||
#include "llama-io.h"
|
||||
#include "llama-graph.h"
|
||||
#include "llama-memory.h"
|
||||
|
||||
#include "ggml-cpp.h"
|
||||
|
||||
#include <set>
|
||||
#include <vector>
|
||||
|
||||
struct llama_cparams;
|
||||
struct llama_hparams;
|
||||
struct llama_ubatch;
|
||||
struct llama_sbatch;
|
||||
struct llama_model;
|
||||
struct llama_context;
|
||||
|
||||
struct llama_kv_cache : public llama_memory_i {
|
||||
virtual ~llama_kv_cache() = default;
|
||||
|
||||
// call if batch processing fails - restores the cache state
|
||||
virtual void restore() = 0;
|
||||
|
||||
// call after successful batch processing - clears any pending state
|
||||
virtual void commit() = 0;
|
||||
|
||||
// process any pending defrag/shift/etc. operations
|
||||
// optionally call once before processing a new batch
|
||||
virtual bool update(llama_context & lctx) = 0;
|
||||
|
||||
// schedule a defrag if the fragmentation threshold is exceeded. otherwise, do nothing
|
||||
virtual void defrag_sched(float thold) = 0;
|
||||
|
||||
// simulate full cache, used for allocating worst-case compute buffers
|
||||
virtual void set_full() = 0;
|
||||
|
||||
//
|
||||
// batch processing
|
||||
//
|
||||
|
||||
virtual llama_sbatch sbatch_init(const llama_batch & batch, bool logits_all) = 0;
|
||||
|
||||
// different KV caches require different batch splitting strategies
|
||||
virtual llama_ubatch ubatch_next(llama_sbatch & sbatch, uint32_t n_ubatch, bool embd_pooled) const = 0;
|
||||
|
||||
// find an empty slot of size "n_tokens" in the cache
|
||||
virtual bool find_slot(const llama_ubatch & batch) = 0;
|
||||
|
||||
// getters
|
||||
virtual int32_t get_n_tokens() const = 0;
|
||||
virtual int32_t get_used_cells() const = 0; // TODO: remove, this is too-specific to the unified cache
|
||||
virtual llama_pos get_pos_max() const = 0;
|
||||
virtual bool get_can_shift() const = 0;
|
||||
|
||||
bool get_can_edit() const override { return get_can_shift(); }
|
||||
|
||||
//
|
||||
// state write/read
|
||||
//
|
||||
|
||||
virtual void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const = 0;
|
||||
virtual void state_read (llama_io_read_i & io, llama_seq_id seq_id = -1) = 0;
|
||||
};
|
||||
|
||||
//
|
||||
// llama_kv_cache_guard
|
||||
//
|
||||
|
||||
struct llama_kv_cache_guard {
|
||||
llama_kv_cache_guard(llama_kv_cache * kv) : kv(kv) {}
|
||||
|
||||
~llama_kv_cache_guard() {
|
||||
kv->restore();
|
||||
}
|
||||
|
||||
void commit() {
|
||||
kv->commit();
|
||||
}
|
||||
|
||||
private:
|
||||
llama_kv_cache * kv;
|
||||
};
|
||||
|
||||
// block of KV slots to move when defragging
|
||||
struct llama_kv_defrag_move {
|
||||
uint32_t src;
|
||||
uint32_t dst;
|
||||
uint32_t len;
|
||||
};
|
||||
|
||||
//
|
||||
// llama_kv_cache_unified
|
||||
//
|
||||
|
||||
// TODO: add notion of max sequences
|
||||
class llama_kv_cache_unified : public llama_kv_cache {
|
||||
public:
|
||||
struct kv_cell {
|
||||
llama_pos pos = -1;
|
||||
llama_pos delta = 0;
|
||||
|
||||
std::set<llama_seq_id> seq_id;
|
||||
|
||||
bool has_seq_id(const llama_seq_id & id) const {
|
||||
return seq_id.find(id) != seq_id.end();
|
||||
}
|
||||
|
||||
bool is_empty() const {
|
||||
return seq_id.empty();
|
||||
}
|
||||
|
||||
bool is_same_seq(const kv_cell & other) const {
|
||||
return seq_id == other.seq_id;
|
||||
}
|
||||
};
|
||||
|
||||
static uint32_t get_padding(const llama_cparams & cparams);
|
||||
|
||||
llama_kv_cache_unified(
|
||||
const llama_model & model,
|
||||
ggml_type type_k,
|
||||
ggml_type type_v,
|
||||
bool v_trans,
|
||||
bool offload,
|
||||
uint32_t kv_size,
|
||||
uint32_t padding);
|
||||
|
||||
~llama_kv_cache_unified() = default;
|
||||
|
||||
//
|
||||
// llama_memory_i
|
||||
//
|
||||
|
||||
void clear() override;
|
||||
|
||||
bool seq_rm (llama_seq_id seq_id, llama_pos p0, llama_pos p1) override;
|
||||
void seq_cp (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) override;
|
||||
void seq_keep(llama_seq_id seq_id) override;
|
||||
void seq_add (llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos delta) override;
|
||||
void seq_div (llama_seq_id seq_id, llama_pos p0, llama_pos p1, int d) override;
|
||||
|
||||
llama_pos seq_pos_max(llama_seq_id seq_id) const override;
|
||||
|
||||
//
|
||||
// llama_kv_cache
|
||||
//
|
||||
|
||||
void restore() override;
|
||||
void commit() override;
|
||||
|
||||
bool update(llama_context & ctx) override;
|
||||
|
||||
void defrag_sched(float thold) override;
|
||||
|
||||
void set_full() override;
|
||||
|
||||
llama_sbatch sbatch_init(const llama_batch & batch, bool logits_all) override;
|
||||
|
||||
llama_ubatch ubatch_next(llama_sbatch & sbatch, uint32_t n_ubatch, bool embd_pooled) const override;
|
||||
|
||||
// updates the cache head
|
||||
// Note: On success, it's important that cache.head points
|
||||
// to the first cell of the slot.
|
||||
bool find_slot(const llama_ubatch & batch) override;
|
||||
|
||||
int32_t get_n_tokens() const override;
|
||||
int32_t get_used_cells() const override;
|
||||
|
||||
// TODO: better data structures to reduce the cost of this operation
|
||||
llama_pos get_pos_max() const override;
|
||||
|
||||
bool get_can_shift() const override;
|
||||
|
||||
// state write/load
|
||||
|
||||
void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const override;
|
||||
void state_read (llama_io_read_i & io, llama_seq_id seq_id = -1) override;
|
||||
|
||||
// Note: The value of head isn't only used to optimize searching
|
||||
// for a free KV slot. llama_decode_impl also uses it, so it
|
||||
// cannot be freely changed after a slot has been allocated.
|
||||
uint32_t head = 0;
|
||||
uint32_t size = 0;
|
||||
uint32_t used = 0; // used cells (i.e. at least one seq_id)
|
||||
|
||||
// computed before each graph build
|
||||
uint32_t n = 0;
|
||||
|
||||
std::vector<kv_cell> cells;
|
||||
|
||||
std::vector<ggml_tensor *> k_l; // per layer
|
||||
std::vector<ggml_tensor *> v_l;
|
||||
|
||||
private:
|
||||
const llama_model & model;
|
||||
const llama_hparams & hparams;
|
||||
|
||||
bool has_shift = false;
|
||||
bool do_defrag = false;
|
||||
|
||||
bool v_trans = true; // the value tensor is transposed
|
||||
bool can_shift = false;
|
||||
|
||||
// required padding
|
||||
uint32_t padding = 1;
|
||||
|
||||
ggml_type type_k = GGML_TYPE_F16;
|
||||
ggml_type type_v = GGML_TYPE_F16;
|
||||
|
||||
std::vector<ggml_context_ptr> ctxs;
|
||||
std::vector<ggml_backend_buffer_ptr> bufs;
|
||||
|
||||
// defrag
|
||||
struct {
|
||||
std::vector<llama_kv_defrag_move> moves;
|
||||
} defrag_info;
|
||||
|
||||
// return true if cells have been moved
|
||||
bool defrag_prepare(int32_t n_max_nodes);
|
||||
|
||||
// commit/restore cache
|
||||
struct slot_range {
|
||||
uint32_t c0 = 0; // note: these are cell indices, not sequence positions
|
||||
uint32_t c1 = 0;
|
||||
};
|
||||
|
||||
// pending cell updates that are not yet committed
|
||||
struct {
|
||||
std::vector<slot_range> ranges;
|
||||
} pending;
|
||||
|
||||
// find how many cells are currently in use
|
||||
uint32_t cell_max() const;
|
||||
|
||||
size_t total_size() const;
|
||||
|
||||
size_t size_k_bytes() const;
|
||||
size_t size_v_bytes() const;
|
||||
|
||||
ggml_tensor * build_rope_shift(
|
||||
const llama_cparams & cparams,
|
||||
ggml_context * ctx,
|
||||
ggml_tensor * cur,
|
||||
ggml_tensor * shift,
|
||||
ggml_tensor * factors,
|
||||
float freq_base,
|
||||
float freq_scale) const;
|
||||
|
||||
llm_graph_result_ptr build_graph_shift(
|
||||
const llama_cparams & cparams,
|
||||
ggml_context * ctx,
|
||||
ggml_cgraph * gf) const;
|
||||
|
||||
llm_graph_result_ptr build_graph_defrag(
|
||||
const llama_cparams & cparams,
|
||||
ggml_context * ctx,
|
||||
ggml_cgraph * gf,
|
||||
const std::vector<llama_kv_defrag_move> & moves) const;
|
||||
|
||||
void state_write_meta(llama_io_write_i & io, const std::vector<std::pair<uint32_t, uint32_t>> & cell_ranges, llama_seq_id seq_id = -1) const;
|
||||
void state_write_data(llama_io_write_i & io, const std::vector<std::pair<uint32_t, uint32_t>> & cell_ranges) const;
|
||||
|
||||
bool state_read_meta(llama_io_read_i & io, uint32_t cell_count, llama_seq_id dest_seq_id = -1);
|
||||
bool state_read_data(llama_io_read_i & io, uint32_t cell_count);
|
||||
};
|
||||
|
||||
//
|
||||
// llama_kv_cache_recurrent
|
||||
//
|
||||
|
||||
class llama_kv_cache_recurrent : public llama_kv_cache {
|
||||
public:
|
||||
struct kv_cell {
|
||||
llama_pos pos = -1;
|
||||
int32_t src = -1; // used to copy states
|
||||
int32_t tail = -1;
|
||||
|
||||
std::set<llama_seq_id> seq_id;
|
||||
|
||||
bool has_seq_id(const llama_seq_id & id) const {
|
||||
return seq_id.find(id) != seq_id.end();
|
||||
}
|
||||
|
||||
bool is_empty() const {
|
||||
return seq_id.empty();
|
||||
}
|
||||
|
||||
bool is_same_seq(const kv_cell & other) const {
|
||||
return seq_id == other.seq_id;
|
||||
}
|
||||
};
|
||||
|
||||
llama_kv_cache_recurrent(
|
||||
const llama_model & model,
|
||||
ggml_type type_k,
|
||||
ggml_type type_v,
|
||||
bool offload,
|
||||
uint32_t kv_size);
|
||||
|
||||
~llama_kv_cache_recurrent() = default;
|
||||
|
||||
//
|
||||
// llama_memory_i
|
||||
//
|
||||
|
||||
void clear() override;
|
||||
|
||||
bool seq_rm (llama_seq_id seq_id, llama_pos p0, llama_pos p1) override;
|
||||
void seq_cp (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) override;
|
||||
void seq_keep(llama_seq_id seq_id) override;
|
||||
void seq_add (llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos delta) override;
|
||||
void seq_div (llama_seq_id seq_id, llama_pos p0, llama_pos p1, int d) override;
|
||||
|
||||
llama_pos seq_pos_max(llama_seq_id seq_id) const override;
|
||||
|
||||
//
|
||||
// llama_kv_cache
|
||||
//
|
||||
|
||||
void restore() override;
|
||||
void commit() override;
|
||||
|
||||
bool update(llama_context & lctx) override;
|
||||
|
||||
void defrag_sched(float thold) override;
|
||||
|
||||
void set_full() override;
|
||||
|
||||
llama_sbatch sbatch_init(const llama_batch & batch, bool logits_all) override;
|
||||
|
||||
llama_ubatch ubatch_next(llama_sbatch & sbatch, uint32_t n_ubatch, bool embd_pooled) const override;
|
||||
|
||||
bool find_slot(const llama_ubatch & batch) override;
|
||||
|
||||
int32_t get_n_tokens() const override;
|
||||
int32_t get_used_cells() const override;
|
||||
|
||||
// TODO: better data structures to reduce the cost of this operation
|
||||
llama_pos get_pos_max() const override;
|
||||
|
||||
bool get_can_shift() const override;
|
||||
|
||||
// TODO: temporary methods - they are not really const as they do const_cast<>, fix this
|
||||
int32_t s_copy(int i) const;
|
||||
float s_mask(int i) const;
|
||||
|
||||
// state write/load
|
||||
|
||||
void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const override;
|
||||
void state_read (llama_io_read_i & io, llama_seq_id seq_id = -1) override;
|
||||
|
||||
// Note: The value of head isn't only used to optimize searching
|
||||
// for a free KV slot. llama_decode_impl also uses it, so it
|
||||
// cannot be freely changed after a slot has been allocated.
|
||||
uint32_t head = 0;
|
||||
uint32_t size = 0;
|
||||
uint32_t used = 0; // used cells (i.e. at least one seq_id)
|
||||
|
||||
// computed before each graph build
|
||||
uint32_t n = 0;
|
||||
|
||||
std::vector<kv_cell> cells;
|
||||
|
||||
std::vector<ggml_tensor *> k_l; // per layer
|
||||
std::vector<ggml_tensor *> v_l;
|
||||
|
||||
private:
|
||||
//const llama_model & model;
|
||||
const llama_hparams & hparams;
|
||||
|
||||
// commit/restore cache
|
||||
// TODO: rework for recurrent cache
|
||||
struct slot_range {
|
||||
uint32_t c0 = 0; // note: these are cell indices, not sequence positions
|
||||
uint32_t c1 = 0;
|
||||
};
|
||||
|
||||
// pending cell updates that are not yet committed
|
||||
struct {
|
||||
std::vector<slot_range> ranges;
|
||||
} pending;
|
||||
|
||||
ggml_type type_k = GGML_TYPE_F16;
|
||||
ggml_type type_v = GGML_TYPE_F16;
|
||||
|
||||
std::vector<ggml_context_ptr> ctxs;
|
||||
std::vector<ggml_backend_buffer_ptr> bufs;
|
||||
|
||||
// find how many cells are currently in use
|
||||
uint32_t cell_max() const;
|
||||
|
||||
size_t total_size() const;
|
||||
|
||||
size_t size_k_bytes() const;
|
||||
size_t size_v_bytes() const;
|
||||
|
||||
void state_write_meta(llama_io_write_i & io, const std::vector<std::pair<uint32_t, uint32_t>> & cell_ranges, llama_seq_id seq_id = -1) const;
|
||||
void state_write_data(llama_io_write_i & io, const std::vector<std::pair<uint32_t, uint32_t>> & cell_ranges) const;
|
||||
|
||||
bool state_read_meta(llama_io_read_i & io, uint32_t cell_count, llama_seq_id dest_seq_id = -1);
|
||||
bool state_read_data(llama_io_read_i & io, uint32_t cell_count);
|
||||
};
|
||||
|
||||
|
||||
//
|
||||
// kv cache view
|
||||
//
|
||||
|
||||
llama_kv_cache_view llama_kv_cache_view_init(const llama_kv_cache & kv, int32_t n_seq_max);
|
||||
|
||||
void llama_kv_cache_view_update(llama_kv_cache_view * view, const llama_kv_cache * kv);
|
||||
Reference in New Issue
Block a user