Add PLM GGUF Conversion & Inference Support

[Si1w]

This PR adds HF->GGUF conversion & inference support for PLM Model PLM-1.8B-Instruct

The Model has already been converted into gguf form with quantized and tested PLM-1.8B-Instruct-gguf, PLM-1.8B-Instruct-id-gguf

The Model Arch is similar with Deepseek V2 and Minicpm3. The key points of the model are:

• Sparse FFN: PLM uses Squared ReLU (up and down projections)
• MLA: PLM uses Multi-head Latent Attention

The details of the model can be seen in the following Paper

PLM: Efficient Peripheral Language Models Hardware-Co-Designed for Ubiquitous Computing

---

• I have read the contributing guidelines

Self-reported review complexity:

• Low
• Medium
• High

[arch-btw]

Tested both the premade gguf and converting the gguf, both work 👍

Looks like it's using the qwen2 tokenizer with the associated chatml prompt template:

llama_model_loader: - kv   0:                       general.architecture str              = plm
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = PLM 1.8B Instruct
llama_model_loader: - kv   3:                           general.finetune str              = Instruct
llama_model_loader: - kv   4:                           general.basename str              = PLM
llama_model_loader: - kv   5:                         general.size_label str              = 1.8B
llama_model_loader: - kv   6:                            general.license str              = mit
llama_model_loader: - kv   7:                            plm.block_count u32              = 32
llama_model_loader: - kv   8:                         plm.context_length u32              = 4096
llama_model_loader: - kv   9:                       plm.embedding_length u32              = 2048
llama_model_loader: - kv  10:                    plm.feed_forward_length u32              = 8192
llama_model_loader: - kv  11:                   plm.attention.head_count u32              = 16
llama_model_loader: - kv  12:                plm.attention.head_count_kv u32              = 16
llama_model_loader: - kv  13:                         plm.rope.freq_base f32              = 100000.000000
llama_model_loader: - kv  14:       plm.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  15:                             plm.vocab_size u32              = 151936
llama_model_loader: - kv  16:                 plm.attention.kv_lora_rank u32              = 512
llama_model_loader: - kv  17:                   plm.attention.key_length u32              = 192
llama_model_loader: - kv  18:                 plm.attention.value_length u32              = 128
llama_model_loader: - kv  19:                   plm.rope.dimension_count u32              = 64
llama_model_loader: - kv  20:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  21:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  22:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  23:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  24:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  25:                tokenizer.ggml.eos_token_id u32              = 151643
llama_model_loader: - kv  26:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  27:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  28:                    tokenizer.chat_template str              = {% for message in messages %}{% if lo...
llama_model_loader: - kv  29:               general.quantization_version u32              = 2
llama_model_loader: - kv  30:                          general.file_type u32              = 15
llama_model_loader: - type  f32:   97 tensors
llama_model_loader: - type q4_K:  176 tensors
llama_model_loader: - type q6_K:   17 tensors

The only small error was a brief switch in language, but that's probably not related to this PR:

> hello!
通往成功

> How are you?
I'm doing well, thank you! How about you? How can I help you today?

convert_hf_to_gguf.py output:

python convert_hf_to_gguf.py /home/test/PLM-1.8B-Instruct --outtype f32
.....
INFO:hf-to-gguf:Set meta model
INFO:hf-to-gguf:Set model parameters
INFO:hf-to-gguf:gguf: context length = 4096
INFO:hf-to-gguf:gguf: embedding length = 2048
INFO:hf-to-gguf:gguf: feed forward length = 8192
INFO:hf-to-gguf:gguf: head count = 16
INFO:hf-to-gguf:gguf: key-value head count = 16
INFO:hf-to-gguf:gguf: rope theta = 100000.0
INFO:hf-to-gguf:gguf: rms norm epsilon = 1e-06
INFO:hf-to-gguf:gguf: file type = 0
INFO:hf-to-gguf:Set model tokenizer
INFO:gguf.vocab:Adding 151387 merge(s).
INFO:gguf.vocab:Setting special token type eos to 151643
INFO:gguf.vocab:Setting special token type pad to 151643
INFO:gguf.vocab:Setting special token type bos to 151643
INFO:gguf.vocab:Setting chat_template to {% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system
You are a helpful assistant<|im_end|>
' }}{% endif %}{{'<|im_start|>' + message['role'] + '
' + message['content'] + '<|im_end|>' + '
'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
' }}{% endif %}
INFO:hf-to-gguf:Set model quantization version
INFO:gguf.gguf_writer:Writing the following files:
INFO:gguf.gguf_writer:/home/test/PLM-1.8B-Instruct/PLM-1.8B-Instruct-F32.gguf: n_tensors = 290, total_size = 7.3G
Writing: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 7.30G/7.30G [00:11<00:00, 643Mbyte/s]
INFO:hf-to-gguf:Model successfully exported to /home/test/PLM-1.8B-Instruct/PLM-1.8B-Instruct-F32.gguf