llama : add Mixtral support (#4406)
* convert : support Mixtral as LLAMA arch * convert : fix n_ff typo * llama : model loading * ggml : sync latest ggml_mul_mat_id * llama : update graph to support MoE * llama : fix cur -> cur_expert * llama : first working version * llama : fix expert weighting in the FFN * ggml : ggml_get_rows support 2D indexing [n_tokens, n_experts] (cpu only) * ggml : add n_as argument to ggml_mul_mat_id * ggml : fix ggml_get_rows to take into account ne02 / ne11 * metal : add more general support for ggml_get_rows + tests * llama : add basic support for offloading moe with CUDA * metal : add/mul/div use general kernel when src1 not cont * metal : reduce the kernel launches for ggml_mul_mat_id * ggml : get_rows : support non-contiguos tensors with gaps, generalize up to 3D * ggml : update get_rows f16 and q * cuda : support non-contiguous src1 in get_rows * llama : offload missing ffn_moe_silu * metal : fix ggml_get_rows to work with non-cont src1 * metal : add indirect mat-vec kernels for all quantization types * llama : do not quantize expert gating tensors * llama : add n_expert and n_expert_used to hparams + change quants * test-backend-ops : add moe test * cuda : fix get_rows when ncols is odd * convert : determine n_ctx correctly * metal : fix ggml_mul_mat_id for F32 * test-backend-ops : make experts more evenly probable (test_moe) * test-backend-ops : cleanup, add moe test for batches * test-backend-ops : add cpy from f32 -> all types test * test-backend-ops : fix dequantize block offset * llama : fix hard-coded number of experts * test-backend-ops : simplify and disable slow tests to avoid CI timeout * test-backend-ops : disable MOE test with thread sanitizer * cuda : fix mul_mat_id with multi gpu * convert : use 1e6 rope_freq_base for mixtral * convert : fix style * convert : support safetensors format * gguf-py : bump version * metal : add cpy f16 -> f32 kernel * metal : fix binary ops for ne10 % 4 != 0 * test-backend-ops : add one more sum_rows test * ggml : do not use BLAS with ggml_mul_mat_id * convert-hf : support for mixtral-instruct (#4428) * convert : typo fix, add additional hyperparameters, use LLaMA arch for Mixtral-instruct * convert : use sentencepiece tokenizer for Mixtral-instruct * convert : make flake8 happy * metal : fix soft_max kernels ref: https://github.com/ggerganov/ggml/pull/621/commits/1914017863d2f9ab8ecc0281cc2a56d683668b92 * metal : limit kernels to not use more than the allowed threads --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> Co-authored-by: Radek Pilar <github@mrkva.eu>
This commit is contained in:
parent
fecac45658
commit
799a1cb13b
14 changed files with 2370 additions and 395 deletions
|
@ -149,6 +149,11 @@ class TensorNameMap:
|
|||
"model.layers.{bid}.ln2", # yi
|
||||
),
|
||||
|
||||
MODEL_TENSOR.FFN_GATE_INP: (
|
||||
"layers.{bid}.feed_forward.gate", # mixtral
|
||||
"model.layers.{bid}.block_sparse_moe.gate", # mixtral
|
||||
),
|
||||
|
||||
# Feed-forward up
|
||||
MODEL_TENSOR.FFN_UP: (
|
||||
"gpt_neox.layers.{bid}.mlp.dense_h_to_4h", # gptneox
|
||||
|
@ -164,11 +169,21 @@ class TensorNameMap:
|
|||
"transformer.h.{bid}.mlp.w1", # qwen
|
||||
),
|
||||
|
||||
MODEL_TENSOR.FFN_UP_EXP: (
|
||||
"layers.{bid}.feed_forward.experts.{xid}.w3", # mixtral
|
||||
"model.layers.{bid}.block_sparse_moe.experts.{xid}.w3", # mixtral
|
||||
),
|
||||
|
||||
# Feed-forward gate
|
||||
MODEL_TENSOR.FFN_GATE: (
|
||||
"model.layers.{bid}.mlp.gate_proj", # llama-hf refact
|
||||
"layers.{bid}.feed_forward.w1", # llama-pth
|
||||
"transformer.h.{bid}.mlp.w2", # qwen
|
||||
"model.layers.{bid}.mlp.gate_proj", # llama-hf refact
|
||||
"layers.{bid}.feed_forward.w1", # llama-pth
|
||||
"transformer.h.{bid}.mlp.w2", # qwen
|
||||
),
|
||||
|
||||
MODEL_TENSOR.FFN_GATE_EXP: (
|
||||
"layers.{bid}.feed_forward.experts.{xid}.w1", # mixtral
|
||||
"model.layers.{bid}.block_sparse_moe.experts.{xid}.w1", # mixtral
|
||||
),
|
||||
|
||||
# Feed-forward down
|
||||
|
@ -185,6 +200,11 @@ class TensorNameMap:
|
|||
"language_model.encoder.layers.{bid}.mlp.dense_4h_to_h", # persimmon
|
||||
),
|
||||
|
||||
MODEL_TENSOR.FFN_DOWN_EXP: (
|
||||
"layers.{bid}.feed_forward.experts.{xid}.w2", # mixtral
|
||||
"model.layers.{bid}.block_sparse_moe.experts.{xid}.w2", # mixtral
|
||||
),
|
||||
|
||||
MODEL_TENSOR.ATTN_Q_NORM: (
|
||||
"language_model.encoder.layers.{bid}.self_attention.q_layernorm",
|
||||
),
|
||||
|
@ -213,11 +233,14 @@ class TensorNameMap:
|
|||
for tensor, keys in self.block_mappings_cfg.items():
|
||||
if tensor not in MODEL_TENSORS[arch]:
|
||||
continue
|
||||
tensor_name = TENSOR_NAMES[tensor].format(bid = bid)
|
||||
self.mapping[tensor_name] = (tensor, tensor_name)
|
||||
for key in keys:
|
||||
key = key.format(bid = bid)
|
||||
self.mapping[key] = (tensor, tensor_name)
|
||||
# TODO: make this configurable
|
||||
n_experts = 8
|
||||
for xid in range(n_experts):
|
||||
tensor_name = TENSOR_NAMES[tensor].format(bid = bid, xid = xid)
|
||||
self.mapping[tensor_name] = (tensor, tensor_name)
|
||||
for key in keys:
|
||||
key = key.format(bid = bid, xid = xid)
|
||||
self.mapping[key] = (tensor, tensor_name)
|
||||
|
||||
def get_type_and_name(self, key: str, try_suffixes: Sequence[str] = ()) -> tuple[MODEL_TENSOR, str] | None:
|
||||
result = self.mapping.get(key)
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue