GLM-4¶

Zhipu's GLM-4 dense decoder-only LLM, ported to pure Keras 3. It uses partial rotary embeddings (only partial_rotary_factor of each head is rotated) with a NeoX-style rope layout, post-norm residuals, and biased QKV projections.

Note the rope trap: GLM and GLM-4 use interleaved partial rope, while GLM-4.5 and later MoE models use the NeoX (half-split) layout. Mixing them silently destroys parity.

Links:

Variants¶

Load any of these with from_weights("<variant>").

Variant	Hub
`glm-4-9b-0414`	`THUDM/GLM-4-9B-0414`
`glm-4-32b-0414`	`THUDM/GLM-4-32B-0414`
`glm-z1-9b-0414`	`THUDM/GLM-Z1-9B-0414`
`glm-z1-32b-0414`	`THUDM/GLM-Z1-32B-0414`

API¶

`Glm4Model`¶

The decoder backbone, no LM head. Returns {"last_hidden_state": (batch, seq, embed_dim)}.

Arg	Default	Meaning
`vocab_size`	`151552`	token vocabulary size
`embed_dim`	`4096`	model width
`num_layers`	`40`	decoder blocks
`num_heads`	`32`	query heads
`num_kv_heads`	`2`	key/value heads (GQA)
`head_dim`	`128`	per-head width
`mlp_dim`	`13696`	MLP inner width
`partial_rotary_factor`	`0.5`	fraction of each head that gets rotated
`norm_eps`	`1.5625e-07`	RMSNorm epsilon
`rope_theta`	`10000.0`	rotary base frequency
`attention_bias`	`True`	add bias terms to the qkv projections
`tie_embeddings`	`False`	reuse the embedding matrix as the LM head

`Glm4Generate`¶

Glm4Model plus a (tied) LM head. Returns {"logits": (batch, seq, vocab_size)} and adds .generate(). Same constructor arguments as Glm4Model.

generate(input_ids, attention_mask=None, max_new_tokens=None,
         eos_token_id=None, sampler=None, seed=None, **prefill_inputs)

Arg	Default	Meaning
`input_ids`	required	`(batch, seq)` token ids
`attention_mask`	`None`	`(batch, seq)` 1 = keep, 0 = padding
`max_new_tokens`	`None`	tokens to generate
`eos_token_id`	`None`	stop token (defaults to the tokenizer's)
`sampler`	`None`	sampling strategy; greedy when unset
`seed`	`None`	seed for stochastic samplers

`Glm4Tokenizer`¶

Tokenizer on the tokenizers backend.

Glm4Tokenizer(hf_id=None, tokenizer_file=None)

Arg	Default	Meaning
`hf_id`	`None`	Hub repo to pull the tokenizer files from
`tokenizer_file`	`None`	explicit path to a `tokenizer.json`

Calling it returns {"input_ids", "attention_mask"}, padded across the batch. It accepts a plain string or a list of strings (a batch). This tokenizer has no chat template, so pass a prompt you have already rendered rather than a message list. Decode with .decode(ids) for one sequence or .batch_decode(ids) for a batch.

End-to-end example¶

Single input¶

import os
os.environ["KERAS_BACKEND"] = "torch"   # or "jax" / "tensorflow"

from kerasformers.models.glm4 import Glm4Generate, Glm4Tokenizer

model = Glm4Generate.from_weights("glm-4-9b-0414")
tokenizer = Glm4Tokenizer.from_weights("glm-4-9b-0414")

inputs = tokenizer("Explain rotary embeddings in one sentence.")
outputs = model.generate(**inputs, max_new_tokens=64)

print(tokenizer.decode(outputs[0]))

Batch¶

Pass a list of strings. The tokenizer pads them and generate runs the batch together:

prompts = [
    "The capital of France is",
    "In one sentence, what is a transformer?",
    "Write a haiku about GPUs.",
]
inputs = tokenizer(prompts)              # {"input_ids": (3, seq), "attention_mask": (3, seq)}
outputs = model.generate(**inputs, max_new_tokens=64)

for text in tokenizer.batch_decode(outputs):
    print(text)

Backbone only¶

from kerasformers.models.glm4 import Glm4Model

backbone = Glm4Model.from_weights("glm-4-9b-0414")
hidden = backbone(inputs)["last_hidden_state"]   # (batch, seq, embed_dim)

Loading from the Hub¶

Any Hub repo with this architecture works via the hf: prefix, including community fine-tunes:

model = Glm4Generate.from_weights("hf:THUDM/GLM-4-9B-0414")

Lower memory¶

Larger checkpoints load in bf16 or weight-only quantized. See quantization.md:

model = Glm4Generate.from_weights(
    "glm-4-9b-0414", quantization="int8", low_memory=True, load_dtype="bfloat16"
)