Unable to generate imatrix for DeekSeek, "KV cache shifting is not supported for this model (--no-context-shift to disable)"

[bartowski1182]

Name and Version

b4273 for Ubuntu

Operating systems

Linux

GGML backends

CPU, CUDA

Hardware

EPYC 7702, 3090

Models

DeekSeek 2.5 1210

https://huggingface.co/deepseek-ai/DeepSeek-V2.5-1210

Problem description & steps to reproduce

When running imatrix, I receive the error:

KV cache shifting is not supported for this model (--no-context-shift to disable)

However this error doesn't go away even if I pass that parameter:

./llama-imatrix -m /models_out/DeepSeek-V2.5-1210-GGUF/DeepSeek-V2.5-1210-Q8_0.gguf -f /training_dir/calibration_datav3.txt --output-file /models_out/DeepSeek-V2.5-1210-GGUF/DeepSeek-V2.5-1210.imatrix -t 120 --no-context-shift

Also tried with flash attention -fa to no avail

First Bad Commit

No response

Relevant log output

llama_new_context_with_model: n_ctx_per_seq (512) < n_ctx_train (163840) -- the full capacity of the model will not be utilized
llama_kv_cache_init:        CPU KV buffer size =  2400.00 MiB
llama_new_context_with_model: KV self size  = 2400.00 MiB, K (f16): 1440.00 MiB, V (f16):  960.00 MiB
llama_new_context_with_model:        CPU  output buffer size =     0.39 MiB
llama_new_context_with_model:      CUDA0 compute buffer size =  1422.00 MiB
llama_new_context_with_model:  CUDA_Host compute buffer size =    81.01 MiB
llama_new_context_with_model: graph nodes  = 4480
llama_new_context_with_model: graph splits = 1080 (with bs=512), 1 (with bs=1)
common_init_from_params: KV cache shifting is not supported for this model (--no-context-shift to disable)'
main : failed to init

[bartowski1182]

Adding LLAMA_EXAMPLE_IMATRIX to --no-context-shift resolves the issue, I assume that's a reasonable solve?