sync : ggml and llama.cpp #2429
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
* tests: add gradient checking to test-backend-ops * remove old comment * reorder includes * adjust SIN/COS parameters * add documentation, use supports_op if possible
test-backend-ops fails because ggml_cont aborts when invoked passing an unsupported type. This commit makes ggml_cont tests pass Signed-off-by: Salvatore Mesoraca <s.mesoraca16@gmail.com>
sin and cos failed test-backend-ops because they tried to dereference a context pointer that is null on dry runs. This commit prevents that segfault. Signed-off-by: Salvatore Mesoraca <s.mesoraca16@gmail.com>
the old code was trying to print a non-existent field (size) and the struct as a whole (which doesn't have a operator<< override defined). Probably a typo happened during refactoring. Signed-off-by: Salvatore Mesoraca <s.mesoraca16@gmail.com>
* Introduce ggml_compute_threadpool - OpenMP functional: check - Vanilla ggml functional: Check - ggml w/threadpool functional: Check - OpenMP no regression: No glaring problems - Vanilla ggml no regression: No glaring problems - ggml w/threadpool no regression: No glaring problems * Minor fixes * fixed use after release bug * fixed a harmless race condition * Fix Android bulid issue * fix more race conditions * fix deadlock for cases where cgraph.n_nodes == 1 and fix --poll case * threadpool: use cpu_get_num_math to set the default number of threadpool threads This way we avoid using E-Cores and Hyperthreaded siblings. * bench: create fresh threadpool for each test For benchmarking it's better to start a fresh pool for each test with the exact number of threads needed for that test. Having larger pools is suboptimal (causes more load, etc). * atomics: always use stdatomics with clang and use relaxed memory order when polling in ggml_barrier This also removes sched_yield() calls from ggml_barrier() to match OpenMP behavior. * threadpool: make polling the default to match openmp behavior All command line args now allow for setting poll to 0 (false). * threadpool: do not wakeup threads in already paused threadpool * fix potential race condition in check_for_work * threadpool: do not create two threadpools if their params are identical * threadpool: reduce pause/resume/wakeup overhead in common cases We now start threadpool in paused state only if we have two. The resume is now implicit (ie new work) which allows for reduced locking and context-switch overhead. * threadpool: add support for hybrid polling poll params (--poll, ...) now specify "polling level", i.e. how aggresively we poll before waiting on cond.var. poll=0 means no polling, 1 means poll for 128K rounds then wait, 2 for 256K rounds, ... The default value of 50 (ie 50x128K rounds) seems like a decent default across modern platforms. We can tune this further as things evolve. * threadpool: reduce the number of barrier required New work is now indicated with an atomic counter that is incremented for each new graph that needs to be computed. This removes the need for extra barrier for clearing the "new_work" and removes the special case for trivial graphs. * threadpool: remove special-casing for disposable threadpools With the efficient hybrid polling there is no need to make disposable pools any different. This simplifies the overall logic and reduces branching. Include n_threads in debug print for disposable threadpool. Declare pause and stop flags as atomic_bool This doesn't actually generate any memory barriers and simply informs the thread sanitizer that these flags can be written & read by different threads without locking. * threadpool: do not clear barrier counters between graphs computes (fixes race with small graphs) This fixes the race condition with very small graphs where the main thread happens to start a new graph while the workers are just about to exit from barriers. * threadpool: use relaxed order for chunk sync Full memory barrier is an overkill for this since each thread works on different chunk * threadpool: remove abort_callback from threadpool state * threadpool: better naming for thread/cpumask releated functions * threadpool: consistent use of int type for n_threads params * threadpool: add support for ggml_threadpool_params_default/init Also removes the need for explicit mask_specified param. all-zero cpumask means use default (usually inherited) cpu affinity mask. * threadpool: move typedef into ggml.h * threadpool: fix apply_priority() function name * threadpool: fix swift wrapper errors due to n_threads int type cleanup * threadpool: enable --cpu-mask and other threadpool related options only if threadpool is enabled * threadpool: replace checks for compute_thread ret code with proper status check * threadpool: simplify threadpool init logic and fix main thread affinity application Most of the init code is now exactly the same between threadpool and openmp. * threadpool: update threadpool resume/pause function names * threadpool: enable openmp by default for now * threadpool: don't forget to free workers state when omp is enabled * threadpool: avoid updating process priority on the platforms that do not require it On Windows we need to change overall process priority class in order to set thread priorities, but on Linux, Mac, etc we do not need to touch the overall process settings. * threadpool: update calling thread prio and affinity only at start/resume This avoids extra syscalls for each graph_compute() * llama-bench: turn threadpool params into vectors, add output headers, etc * llama-bench: add support for cool off between tests --delay This helps for long running tests on platforms that are thermally limited (phones, laptops, etc). --delay (disabled by default) introduces the sleep for N seconds before starting each test. * threadpool: move process priority setting into the apps (bench and cli) This avoids changing the overall process priority on Windows for the apps that use ggml/llama.cpp directy. * threadpool: move all pause/resume logic into ggml * threadpool: futher api cleanup and prep for future refactoring All threadpool related functions and structs use ggml_threadpool prefix. * threadpool: minor indent fixes * threadpool: improve setprioty error message * Update examples/llama-bench/llama-bench.cpp Co-authored-by: slaren <slarengh@gmail.com> * threadpool: fix indent in set_threadpool call * use int32_t for n_thread type in public llama.cpp API * threadpool: use _new and _free instead of _create and _release * fix two more public APIs to use int32_t for n_threads * build: set _GNU_SOURCE for Adroid --------- Co-authored-by: Max Krasnyansky <quic_maxk@quicinc.com> Co-authored-by: fmz <quic_fzaghlou@quic.com> Co-authored-by: Max Krasnyansky <max.krasnyansky@gmail.com> Co-authored-by: slaren <slarengh@gmail.com>
* convert_hf_to_gguf: Add support for RWKV v6 Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Add RWKV tokenization * Fix build Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Do not use special tokens when matching in RWKV tokenizer * Fix model loading * Add (broken) placeholder graph builder for RWKV * Add workaround for kv cache * Add logits conversion to rwkv5 * Add rwkv5 layer norms * Add time mix KVRG & correct merge mistake * Add remaining time mix parameters * Add time mix output loading * Add placeholder llm_build_time_mix * Fix build Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Load more tensors for rwkv v6 Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Fix rwkv tokenizer Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * ggml: Add unary operator Exp Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * RWKV v6 graph building Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Add ``rescale_every_n_layers`` parameter Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Add ``wkv.head_size`` key for RWKV so it doesn't reuse Mamba ssm parameters Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Fix offloading layers to CUDA Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Fix parallel inferencing for RWKV Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Remove trailing whitespaces Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * build_rwkv: Avoid using inplace operations Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * convert_hf_to_gguf: rwkv: Avoid using ``eval`` Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * convert_hf_to_gguf: rwkv tokenizer: Don't escape sequences manually Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Update convert_hf_to_gguf.py Co-authored-by: compilade <git@compilade.net> * ggml: Add backward computation for unary op ``exp`` Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Update convert_hf_to_gguf.py Co-authored-by: compilade <git@compilade.net> * Update convert_hf_to_gguf.py Co-authored-by: compilade <git@compilade.net> * Use MODEL_ARCH.RWKV6 instead of MODEL_ARCH.RWKV Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * build_rwkv6: Simplify graph Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Detect model.type Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Fix tensor loading for 7B/14B models Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Fix group_norm assertion failure with Metal Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Clean up Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Add quantization tensor exclusion Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Use the new advanced batch splits Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * Update src/llama.cpp Co-authored-by: compilade <git@compilade.net> * llama: rwkv6: Use ``ggml_norm`` instead of ``ggml_group_norm`` Co-authored-by: compilade <git@compilade.net> * llama: rwkv6: Apply code style and misc changes Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * converter: Use class name ``Rwkv6Model`` Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Make use of key ``feed_forward_length`` Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Add kv ``time_mix_extra_dim`` and ``time_decay_extra_dim`` Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * converter: Match ``new_name`` instead of ``name`` for float32 explicit tensors Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Keep ``time_mix_w1/w2`` as F32 Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Remove unused nodes Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Apply code format changes Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * llama: rwkv6: Add lora for some supported tensors Currently att.key/receptance/value/gate/output, ffn.receptance/key/value, as well as head.weight Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * rwkv : speed-up tokenization using trie * minor : style + indentation * llama: rwkv6: Avoid division by zero Co-authored-by: compilade <git@compilade.net> * ggml: rwkv_wkv: Avoid copying the state Signed-off-by: Molly Sophia <mollysophia379@gmail.com> --------- Signed-off-by: Molly Sophia <mollysophia379@gmail.com> Co-authored-by: Layl Bongers <3094382+LaylBongers@users.noreply.github.com> Co-authored-by: compilade <git@compilade.net> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Fixed dmmv dequant for ncols== GGML_SYCL_DMMV_X
* Add AVX2 based implementations for quantize_q8_0_4x8, ggml_gemv_q4_0_8x8_q8_0 and ggml_gemm_q4_0_8x8_q8_0 functions * Update code to fix issues occuring due to non alignment of elements to be processed as multiple of 16 in MSVC * Update comments and indentation * Make updates to reduce number of load instructions
* ggml-quants : 1.625 bpw ternary packing for BitNet 1.58b * ggml-quants : faster 1.625 bpw AVX2 vec_dot Not using a lookup table anymore makes it match q4_0 speed. * gguf-py : fix formatting * llama : remove spaces on empty line * ggml-quants : subtract 1 when back in epi8 This makes the 1.625 bpw type go faster than q4_0. Still not the fastest. * ggml-quants : Q2_2 now faster than Q4_K on with AVX2 * ggml-quants : cleanup Q1_3 code formatting * ggml-quants : ARM NEON vec_dot for q2_2 and q1_3 * ggml-quants : use ceiling division when quantizing q1_3 * convert-hf : simplify BitNet pre-quantization This still results in the exact same tensor weights and scales, but it reveals some weirdness in the current algorithm. * convert-hf : allow converting the weird BitNet 1.3B Its FFN size is 5460 which is not convenient. The offending tensors are kept in F16, which makes the final model 5.01 bpw. * bitnet : replace 1.58b with b1.58, as in the paper * ggml-quants : fix build failure on Windows * ggml-quants : attempt to fix Arm 32-bit support * ggml : add some informative comments in q1_3 vec_dot * ggml : add TQ1_0 and TQ2_0 ternary quantization types * ggml : even faster TQ2_0 * ggml : also faster TQ1_0 Same optimization as for TQ2_0 by offsetting the sum instead of the weights. This makes TQ1_0 almost as fast as Q8_0 on AVX2. * ggml : fix build issues in certain environments * ggml : add NEON vec_dot implementation for TQ1_0 and TQ2_0 * ggml : avoid directly using vmlal_high_s8, for 32-bit ARM compat The compiler seems smart enough to use the same instruction even when using vget_high_s8 instead. * ggml : remove q1_3 and q2_2 No more 1.625 bpw and 2.000 bpw, now instead using 1.6875 bpw and 2.0625 bpw with TQ1_0 and TQ2_0, respectively. * llama : remove the separate scale tensors of BitNet b1.58 They won't be needed, since the remaining ternary quant types have built-in scales. * ggml-quants : rename fields of TQ1_0 and TQ2_0 structs for consistency * ggml-quants : allow using vdotq_s32 in TQ2_0 vec_dot Not yet tested on hardware which supports it, might not work or might not even compile. But also it might. It should make the performance better on recent ARM CPUs. * ggml-quants : remove comment about possible format change of TQ2_0 Making it slightly more convenient for AVX512 but less convenient for everything else is not worth the trouble. * gguf-py : Numpy (de)quantization for TQ1_0 and TQ2_0 * ggml-quants : use roundf instead of nearest_int for TQ1_0 and TQ2_0 This does not change anything for ternary models, since their values should never end up being in halfway cases anyway. * convert : allow direct conversion to TQ1_0 and TQ2_0 The token embeddings and output tensors are kept in F16 to allow quantizing them to Q4_K and Q6_K with llama-quantize. * llama : handle fallback for TQ1_0 and TQ2_0 with Q4_0 Q4_0 is not completely symmetric (so not lossless for ternary models), but it should be good enough. * ggml-quants : allow using ARM dot product instructions for TQ1_0 * ggml-quants : deduplicate TQ1_0 and TQ2_0 __ARM_FEATURE_DOTPROD support * ggml : remove unused ggml_mul special case It would otherwise conflict with the more general optimization coming with Mamba-2. * ggml : handle TQ1_0 and TQ2_0 in dequantization-based operators * test-backend-ops : add TQ1_0 and TQ2_0 comments for later Not yet adding uncommented, because some backends like SYCL and Metal do not properly handle unknown types in supports_op for GGML_OP_MUL_MAT. (and Metal also doesn't handle it with GGML_OP_GET_ROWS) Support for TQ1_0 and TQ2_0 for other backends than CPU will be added in follow-up pull requests.
* Improve Vulkan shader builds system - Add dependency to vulkan-shaders-gen to rebuild shaders when changing the shader compilation utility. - Add option to generate debug info for Vulkan shaders to provide shader source to Vulkan shader profiling tools * remove not required self dependency
* ggml : fix missing cpu_set_t on emscripten * better version * bring back android part
* CUDA eval works * stochastic gradient descent op * Adam except decay * CUDA CROSS_ENTROPY_LOSS_BACK * CUDA mnist-fc training works * backend CLI arg * refactor gguf load * remove sched from opt_step_adam * implement l1 regularization (weight decay) * extra call to add optimizer * initialize gradients with ggml_graph_reset * gradient accumulation * increment iter per eval instead of epoch * adjust backend interfaces * fix ggml_graph_reset without backend * fix ggml graph export/import * fixup * rename * revert ggml_opt changes * more general CUDA repeat_back * update documentation, fix CNN * validation split * add clarifying comment * optimize PyTorch training * adjust buffer size, thread count * fix 0.0f validation split * Update examples/mnist/mnist-common.cpp Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * fix gradient accumulation * tensor flag for accumulators -> tensor hash set * Update include/ggml.h Co-authored-by: slaren <slarengh@gmail.com> * Update tests/test-backend-ops.cpp Co-authored-by: slaren <slarengh@gmail.com> * Update tests/test-backend-ops.cpp Co-authored-by: slaren <slarengh@gmail.com> * fix test prints * Update src/ggml-backend.c Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * better CUDA support for noncontiguous out_prod * add comment --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> Co-authored-by: slaren <slarengh@gmail.com>
* add check malloc result on device * update for review comments, check all malloc_device() result --------- Co-authored-by: arthw <14088817+arthw@users.noreply.github.com>
…by submitting smaller cmdbuffers early. (llama/9118) * Overlap cmdbuffer creation and cmdbuffer execution in Vulkan backend by submitting smaller cmdbuffers early. * fix compile issues * Fix issues where the last submit wasn't executed or handled properly. * remove trailing whitespace * Repair GGML_VULKAN_CHECK_RESULTS * Increase submit counter only if actual work has been submitted and increase submit count to 100. * Fix some nodes are not checked with GGML_VULKAN_CHECK_RESULTS enabled.
* Implemented vector length agnostic SVE using switch case for 512-bit, 256-bit, 128-bit vector lengths * Implemented vector length agnostic SVE using switch case for 512-bit, 256-bit, 128-bit vector lengths * Removed WhiteSpaces * ggml : style changes + fix 512-bit nb loop check - fix local scope in switch cases - consistent predicate names - empty lines when necessary - opening braces, spaces - const-correctness - add asserts * Update ggml/src/ggml-quants.c Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* rpc : fix nkvo * rpc : buf_size must not be static ref: #9337 --------- Co-authored-by: slaren <slarengh@gmail.com>
* sycl : update support condition to im2col Signed-off-by: Alberto Cabrera <alberto.cabrera@codeplay.com> * Added TODO to remind supporting FP32 im2col --------- Signed-off-by: Alberto Cabrera <alberto.cabrera@codeplay.com>
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
…/9442) - Added ggml_cpu_has_riscv_v() in GGML to print system info in log - Modified Makefile to only use flag when cross compiling for RISC-V
* feat: Add host buffer type for Ascend NPU(CANN backend) * fix some checking errors * Add a few comments
* cmake : do not hide GGML options ggml-ci * build : rename flag GGML_CUDA_USE_GRAPHS -> GGML_CUDA_GRAPHS for consistency ggml-ci
* threadpool: skip polling for unused threads Currently all threads do N polling rounds even if only 1 thread is active (n_threads_cur == 1). This commit adds a check to skip the polling for unused threads (ith >= n_threads_cur). n_threads_cur is now an atomic_int to explicitly tell thread sanitizer that it is written from one thread and read from other threads (not a race conditions). * threadpool: further simplify and improve ggml_barrier Avoid using strict memory order while polling, yet make sure that all threads go through full memory barrier (memory fence) on ggml_barrier entrace and exit. * threads: add simple barrier test This test does lots of small, parallel matmul ops where the barriers in between dominate the overhead. * threadpool: improve thread sync for new-graphs Using the same tricks as ggml_barrier. All the polling is done with relaxed memory order to keep it efficient, once the new graph is detected we do full fence using read-modify-write with strict memory order. * threadpool: improve abort handling Do not use threadpool->ec (exit code) to decide whether to exit the compute loop. threadpool->ec is not atomic which makes thread-sanitizer rightfully unhappy about it. Instead introduce atomic threadpool->abort flag used for this. This is consistent with how we handle threadpool->stop or pause. While at it add an explicit atomic_load for n_threads_cur for consistency. * test-barrier: release threadpool before releasing the context fixes use-after-free detected by gcc thread-sanitizer on x86-64 for some reason llvm sanitizer is not detecting this issue.
* ggml : fix n_threads_cur initialization with one thread * Update ggml/src/ggml.c --------- Co-authored-by: Max Krasnyansky <quic_maxk@quicinc.com>
ggml-ci
- d6a04f87 - 23e0d70b
* Avoid using saved CUDA graph if scale changes and reset nodes/params on update Fixes ggerganov/llama.cpp#9451 * clear before resize
* ggml: CUDA unary op EXP Signed-off-by: Molly Sophia <mollysophia379@gmail.com> * ggml: rwkv_wkv op CUDA impl Signed-off-by: Molly Sophia <mollysophia379@gmail.com> --------- Signed-off-by: Molly Sophia <mollysophia379@gmail.com>
Signed-off-by: Molly Sophia <mollysophia379@gmail.com>
…e Flash Attention on QY1 (MTT S80) (llama/9526) * mtgpu: add mp_21 support Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com> * mtgpu: disable flash attention on qy1 (MTT S80); disable q3_k and mul_mat_batched_cublas Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com> * mtgpu: enable unified memory Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com> * mtgpu: map cublasOperation_t to mublasOperation_t (sync code to latest) Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com> --------- Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
This reverts commit 50addec9a532a6518146ab837a85504850627316.
* AVX512 version of ggml_gemm_q4_0_8x8_q8_0 * Remove zero vector parameter passing * Rename functions and rearrange order of macros * Edit commments * style : minor adjustments * Update x to start from 0 --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
…lama/9598) Make sure n_barrier and n_barrier_passed do not share the cache line to avoid cache line bouncing. This optimization shows performance improvements even for n_threads <= 8 cases. Resurect TSAN (Thread Sanitizer) check so that we can avoid doing expensive read-modify-write in the normal case and just use thread-fence as originally intended.
llama: enable K-shift for quantized KV cache It will fail on unsupported backends or quant types.
We're missing atomic_thread_fence() in MSVC builds when openmp is disabled.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.