diff options
| author | Pierrick Hymbert <pierrick.hymbert@gmail.com> | 2024-04-11 14:51:07 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-04-11 14:51:07 +0200 |
| commit | b804b1ef77351d2a11be945462c6c251710476cb (patch) | |
| tree | f963c03b90a54083ee67c22c882d20e388820897 /examples/eval-callback/README.md | |
| parent | 8228b66dbc16290c5cbd70e80ab47c068e2569d8 (diff) | |
eval-callback: Example how to use eval callback for debugging (#6576)
* gguf-debug: Example how to use ggml callback for debugging
* gguf-debug: no mutex, verify type, fix stride.
* llama: cv eval: move cb eval field in common gpt_params
* ggml_debug: use common gpt_params to pass cb eval.
Fix get tensor SIGV random.
* ggml_debug: ci: add tests
* ggml_debug: EOL in CMakeLists.txt
* ggml_debug: Remove unused param n_batch, no batching here
* ggml_debug: fix trailing spaces
* ggml_debug: fix trailing spaces
* common: fix cb_eval and user data not initialized
* ci: build revert label
* ggml_debug: add main test label
* doc: add a model: add a link to ggml-debug
* ggml-debug: add to make toolchain
* ggml-debug: tests add the main label
* ggml-debug: ci add test curl label
* common: allow the warmup to be disabled in llama_init_from_gpt_params
* ci: add curl test
* ggml-debug: better tensor type support
* gitignore : ggml-debug
* ggml-debug: printing also the sum of each tensor
* ggml-debug: remove block size
* eval-callback: renamed from ggml-debug
* eval-callback: fix make toolchain
---------
Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Diffstat (limited to 'examples/eval-callback/README.md')
| -rw-r--r-- | examples/eval-callback/README.md | 95 |
1 files changed, 95 insertions, 0 deletions
diff --git a/examples/eval-callback/README.md b/examples/eval-callback/README.md new file mode 100644 index 00000000..66a37e87 --- /dev/null +++ b/examples/eval-callback/README.md @@ -0,0 +1,95 @@ +# llama.cpp/examples/eval-callback + +A simple example which demonstrates how to use callback during the inference. +It simply prints to the console all operations and tensor data. + +Usage: + +```shell +eval-callback \ + --hf-repo ggml-org/models \ + --hf-file phi-2/ggml-model-q4_0.gguf \ + --model phi-2-q4_0.gguf \ + --prompt hello \ + --seed 42 \ + -ngl 33 +``` + +Will print: + +```shell +llm_load_tensors: offloaded 33/33 layers to GPU +... +llama_new_context_with_model: n_ctx = 512 +... +llama_new_context_with_model: CUDA0 compute buffer size = 105.00 MiB +llama_new_context_with_model: CUDA_Host compute buffer size = 6.01 MiB +llama_new_context_with_model: graph nodes = 1225 +llama_new_context_with_model: graph splits = 2 +ggml_debug: inp_embd = (f32) GET_ROWS(token_embd.weight{2560, 51200, 1, 1}, inp_tokens{1, 1, 1, 1}}) = {2560, 1, 1, 1} + [ + [ + [ -0.0181, 0.0272, 0.0272, ...], + ], + ] +ggml_debug: norm-0 = (f32) NORM(CUDA0#inp_embd#0{2560, 1, 1, 1}, }) = {2560, 1, 1, 1} + [ + [ + [ -0.6989, 1.0636, 1.0636, ...], + ], + ] +ggml_debug: norm_w-0 = (f32) MUL(norm-0{2560, 1, 1, 1}, blk.0.attn_norm.weight{2560, 1, 1, 1}}) = {2560, 1, 1, 1} + [ + [ + [ -0.1800, 0.2817, 0.2632, ...], + ], + ] +ggml_debug: attn_norm-0 = (f32) ADD(norm_w-0{2560, 1, 1, 1}, blk.0.attn_norm.bias{2560, 1, 1, 1}}) = {2560, 1, 1, 1} + [ + [ + [ -0.1863, 0.2970, 0.2604, ...], + ], + ] +ggml_debug: wqkv-0 = (f32) MUL_MAT(blk.0.attn_qkv.weight{2560, 7680, 1, 1}, attn_norm-0{2560, 1, 1, 1}}) = {7680, 1, 1, 1} + [ + [ + [ -1.1238, 1.2876, -1.8086, ...], + ], + ] +ggml_debug: bqkv-0 = (f32) ADD(wqkv-0{7680, 1, 1, 1}, blk.0.attn_qkv.bias{7680, 1, 1, 1}}) = {7680, 1, 1, 1} + [ + [ + [ -1.1135, 1.4604, -1.9226, ...], + ], + ] +ggml_debug: bqkv-0 (view) = (f32) VIEW(bqkv-0{7680, 1, 1, 1}, }) = {2560, 1, 1, 1} + [ + [ + [ -1.1135, 1.4604, -1.9226, ...], + ], + ] +ggml_debug: Qcur-0 = (f32) CONT(bqkv-0 (view){2560, 1, 1, 1}, }) = {2560, 1, 1, 1} + [ + [ + [ -1.1135, 1.4604, -1.9226, ...], + ], + ] +ggml_debug: Qcur-0 (reshaped) = (f32) RESHAPE(Qcur-0{2560, 1, 1, 1}, }) = {80, 32, 1, 1} + [ + [ + [ -1.1135, 1.4604, -1.9226, ...], + [ -0.3608, 0.5076, -1.8866, ...], + [ 1.7643, 0.0273, -2.1065, ...], + ... + ], + ] +ggml_debug: Qcur-0 = (f32) ROPE(Qcur-0 (reshaped){80, 32, 1, 1}, CUDA0#inp_pos#0{1, 1, 1, 1}}) = {80, 32, 1, 1} + [ + [ + [ -1.1135, 1.4604, -1.9226, ...], + [ -0.3608, 0.5076, -1.8866, ...], + [ 1.7643, 0.0273, -2.1065, ...], + ... + ], + ] +``` |