Fix incorrect vocab size retrieval in GGUF config

[Isotr0py]

What does this PR do?

Fixes #32526

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[Isotr0py]

cc @kunger97 @SunMarc

[kunger97]

worked on 13B and 32B

[ArthurZucker]

Cool! Is there a reason why it's missing from the config? (or why we don't take it from the gguf file?)

[Isotr0py]

Because there is an issue in llama.cpp's convert_hf_to_gguf.py script. The problem is that they only called gguf_writer.add_vocab_size for llama so that the other architecture won't add vocab_size to the gguf file.

For example, this is the llama convert code with self.gguf_writer.add_vocab_size presented:

    def set_gguf_parameters(self):
        super().set_gguf_parameters()
        hparams = self.hparams
        self.gguf_writer.add_vocab_size(hparams["vocab_size"])

        ...

        # Apply to granite small models only
        if self.hparams.get("vocab_size", 32000) == 49152:
            self.gguf_writer.add_add_bos_token(False)

And the base set_gguf_parameters used in Qwen2 convert which no self.gguf_writer.add_vocab_size presented:

    def set_gguf_parameters(self):
        self.gguf_writer.add_block_count(self.block_count)

        if (n_ctx := self.find_hparam(["max_position_embeddings", "n_ctx"], optional=True)) is not None:
            self.gguf_writer.add_context_length(n_ctx)
            logger.info(f"gguf: context length = {n_ctx}")

        n_embd = self.find_hparam(["hidden_size", "n_embd"])
        self.gguf_writer.add_embedding_length(n_embd)
        logger.info(f"gguf: embedding length = {n_embd}")

        if (n_ff := self.find_hparam(["intermediate_size", "n_inner"], optional=True)) is not None:
            self.gguf_writer.add_feed_forward_length(n_ff)
            logger.info(f"gguf: feed forward length = {n_ff}")

        n_head = self.find_hparam(["num_attention_heads", "n_head"])
        self.gguf_writer.add_head_count(n_head)
        logger.info(f"gguf: head count = {n_head}")

        if (n_head_kv := self.hparams.get("num_key_value_heads")) is not None:
            self.gguf_writer.add_head_count_kv(n_head_kv)
            logger.info(f"gguf: key-value head count = {n_head_kv}")

        if (rope_theta := self.hparams.get("rope_theta")) is not None:
            self.gguf_writer.add_rope_freq_base(rope_theta)
            logger.info(f"gguf: rope theta = {rope_theta}")
        if (f_rms_eps := self.hparams.get("rms_norm_eps")) is not None:
            self.gguf_writer.add_layer_norm_rms_eps(f_rms_eps)
            logger.info(f"gguf: rms norm epsilon = {f_rms_eps}")
        if (f_norm_eps := self.find_hparam(["layer_norm_eps", "layer_norm_epsilon", "norm_epsilon"], optional=True)) is not None:
            self.gguf_writer.add_layer_norm_eps(f_norm_eps)
            logger.info(f"gguf: layer norm epsilon = {f_norm_eps}")
        if (n_experts := self.hparams.get("num_local_experts")) is not None:
            self.gguf_writer.add_expert_count(n_experts)
            logger.info(f"gguf: expert count = {n_experts}")
        if (n_experts_used := self.hparams.get("num_experts_per_tok")) is not None:
            self.gguf_writer.add_expert_used_count(n_experts_used)
            logger.info(f"gguf: experts used count = {n_experts_used}")

        if (head_dim := self.hparams.get("head_dim")) is not None:
            self.gguf_writer.add_key_length(head_dim)
            self.gguf_writer.add_value_length(head_dim)

        self.gguf_writer.add_file_type(self.ftype)
        logger.info(f"gguf: file type = {self.ftype}")

[SunMarc]

Thanks for fixing !

[ArthurZucker]

Got it thanks. For posterity can you add a small comment about the gguf conversion script not adding the vocab size 🤗

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[ArthurZucker]

Thanks 🎈