convert : fix RWKV v6 model conversion (#10913)

* Enable --no-context-shift for llama-perplexity example

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

* RWKV 6: Fix error in ggml_cuda_op_bin_bcast

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>

---------

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>
This commit is contained in:
Molly Sophia 2024-12-20 17:44:58 +08:00 committed by GitHub
parent d408bb9268
commit 0a11f8b7b5
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 4 additions and 1 deletions

View File

@ -626,7 +626,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
[](common_params & params) {
params.ctx_shift = false;
}
).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_IMATRIX}).set_env("LLAMA_ARG_NO_CONTEXT_SHIFT"));
).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_PERPLEXITY}).set_env("LLAMA_ARG_NO_CONTEXT_SHIFT"));
add_opt(common_arg(
{"--chunks"}, "N",
string_format("max number of chunks to process (default: %d, -1 = all)", params.n_chunks),

View File

@ -3065,6 +3065,9 @@ class Rwkv6Model(Model):
if new_name.endswith("time_mix_w2.weight"):
data_torch = data_torch.permute(0, 2, 1)
if new_name.endswith("time_mix_decay.weight") or "lerp" in new_name:
data_torch = data_torch.squeeze()
rescale_every_n_layers = self.hparams["rescale_every"]
if rescale_every_n_layers > 0:
if new_name.endswith("time_mix_output.weight") or new_name.endswith("channel_mix_value.weight"):