Cohere 2 (Command R7B)¶

Cohere's second-generation decoder-only LLM, ported to pure Keras 3. It keeps Cohere's mean-centered LayerNorm, parallel attention/MLP and logit_scale, and adds alternating sliding-window and global attention layers (sliding_window_pattern).

Links:

HF docs: transformers/model_doc/cohere2

Variants¶

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

Variant	Hub
`c4ai-command-r7b-12-2024`	`CohereLabs/c4ai-command-r7b-12-2024`
`c4ai-command-r7b-arabic-02-2025`	`CohereLabs/c4ai-command-r7b-arabic-02-2025`
`command-a-03-2025`	`CohereLabs/c4ai-command-a-03-2025`
`command-a-reasoning-08-2025`	`CohereLabs/command-a-reasoning-08-2025`
`command-a-translate-08-2025`	`CohereLabs/command-a-translate-08-2025`
`tiny-aya-base`	`CohereLabs/tiny-aya-base`
`tiny-aya-earth`	`CohereLabs/tiny-aya-earth`
`tiny-aya-global`	`CohereLabs/tiny-aya-global`
`tiny-aya-water`	`CohereLabs/tiny-aya-water`

API¶

`Cohere2Model`¶

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

Arg	Default	Meaning
`vocab_size`	`256000`	token vocabulary size
`embed_dim`	`8192`	model width
`num_layers`	`40`	decoder blocks
`num_heads`	`64`	query heads
`num_kv_heads`	`64`	key/value heads (GQA)
`head_dim`	`None`	per-head width
`mlp_dim`	`22528`	MLP inner width
`sliding_window`	`4096`	local attention span
`sliding_window_pattern`	`4`	one global layer every N
`norm_eps`	`1e-05`	RMSNorm epsilon
`rope_theta`	`10000.0`	rotary base frequency
`attention_bias`	`False`	add bias terms to the qkv projections
`logit_scale`	`0.0625`	output logit scaling
`tie_embeddings`	`True`	reuse the embedding matrix as the LM head
`layer_types`	`None`

`Cohere2Generate`¶

Cohere2Model plus a (tied) LM head. Returns {"logits": (batch, seq, vocab_size)} and adds .generate(). Same constructor arguments as Cohere2Model.

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

`Cohere2Tokenizer`¶

Tokenizer on the tokenizers backend.

Cohere2Tokenizer(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.cohere2 import Cohere2Generate, Cohere2Tokenizer

model = Cohere2Generate.from_weights("c4ai-command-r7b-12-2024")
tokenizer = Cohere2Tokenizer.from_weights("c4ai-command-r7b-12-2024")

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.cohere2 import Cohere2Model

backbone = Cohere2Model.from_weights("c4ai-command-r7b-12-2024")
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 = Cohere2Generate.from_weights("hf:CohereLabs/c4ai-command-r7b-12-2024")

Lower memory¶

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

model = Cohere2Generate.from_weights(
    "c4ai-command-r7b-12-2024", quantization="int8", low_memory=True, load_dtype="bfloat16"
)