gm

[[Source]]

from gemma import gm

Kauldron API for Gemma.

All symbols

Module

gm	Kauldron API for Gemma.
gm.ckpts	Checkpoints API.
gm.data	Data pipeline ops.
gm.evals	Evaluators for Gemma.
gm.losses	Losses.
gm.math	Math utils (attention masks, positional embeddings, …).
gm.nn	Gemma models.
gm.nn.config	Symbols needed to build new TransformerConfig.
gm.sharding	Sharding utilities.
gm.testing	Testing utilities (dummy models, tokenizer,…).
gm.text	Text processing utilities.
gm.tools	Tools.
gm.typing	Common types for Gemma.

Class

gm.ckpts.AnchoredPolicyLoader	Loader for gm.nn.AnchoredPolicy models.
gm.ckpts.CheckpointPath	Hardcoded paths to Gemma checkpoints.
gm.ckpts.LoadCheckpoint	Loads weights from a Gemma checkpoint.
gm.ckpts.SkipLoRA	Wraps a partial loader to not restore the LoRA weights.
gm.data.AddSeq2SeqFields	Adds the model input, target and loss_mask.
gm.data.ContrastiveTask	Creates the contrastive model inputs for DPO-like loss.
gm.data.DecodeBytes	Decode bytes to str.
gm.data.FormatText	Equivalent to template.format(text=my_string).
gm.data.MapInts	Replace each int by a new value.
gm.data.Pad	Add zeros to the end of the sequence to reach the max length.
gm.data.Parquet	Parquet(*, _fake_refs: ‘type[_FakeRefsUnset]
gm.data.Seq2SeqTask	Sequence-to-sequence task.
gm.data.Tokenize	Tokenize a string to ids.
gm.evals.SamplerEvaluator	Sampling evaluator.
gm.losses.DpoLoss	DPO loss.
gm.losses.NpoLoss	NPO loss.
gm.nn.AnchoredPolicy	Wrapper around a model to compute policy and anchor outputs.
gm.nn.AnchoredPolicyOutput	Output of the gm.nn.AnchoredPolicy.
gm.nn.Attention	Attention module.
gm.nn.AttentionType
gm.nn.Block	Transformer block.
gm.nn.Einsum	Einsum is a convenience module for parameterized tensor multiplication.
gm.nn.Embedder	Embedder module.
gm.nn.FeedForward	Feed forward module.
gm.nn.Gemma2_27B	Gemma2 transformer architecture.
gm.nn.Gemma2_2B	Gemma2 transformer architecture.
gm.nn.Gemma2_9B	Gemma2 transformer architecture.
gm.nn.Gemma3_12B	Gemma3 transformer architecture.
gm.nn.Gemma3_1B	Gemma3 transformer architecture.
gm.nn.Gemma3_270M	Gemma3 transformer architecture.
gm.nn.Gemma3_27B	Gemma3 transformer architecture.
gm.nn.Gemma3_4B	Gemma3 transformer architecture.
gm.nn.Gemma3n_E2B	Gemma3n E2B transformer architecture.
gm.nn.Gemma3n_E4B	Gemma3n E4B transformer architecture.
gm.nn.Gemma4_26B_A4B	Gemma 4 26B_A4B MoE model.
gm.nn.Gemma4_31B	Gemma 4 31B model.
gm.nn.Gemma4_E2B	Gemma 4 E2B model.
gm.nn.Gemma4_E4B	Gemma 4 E4B model.
gm.nn.IntWrapper	Wrapper around a Gemma model to enable int4 inference.
gm.nn.LoRA	Wrapper around a Gemma model to enable LoRA.
gm.nn.Output	Output of the Gemma model.
gm.nn.QuantizationAwareWrapper	Wrapper around a Gemma model to enable quantization aware training.
gm.nn.RMSNorm	RMSNorm layer.
gm.nn.SigLiPFromPatches	SigLIP vision encoder forward pass from PatchifiedMedia.
gm.nn.Transformer	Base transformer class.
gm.nn.TransformerLike	Protocol for a transformer model to be used with a Sampler.
gm.testing.DummyGemma	Dummy transformer architecture, for testing.
gm.testing.DummyTokenizer	Dummy tokenizer.
gm.text.ChatSampler	Chat sampler.
gm.text.Gemma2Tokenizer	Tokenizer for Gemma 2.
gm.text.Gemma3Tokenizer	Tokenizer for Gemma 3.
gm.text.Gemma3nTokenizer	Tokenizer for Gemma3n.
gm.text.Gemma4Sampler	Stateless sampler for Gemma4 with variable-aspect-ratio image support.
gm.text.Gemma4Tokenizer	Tokenizer for Gemma 4.
gm.text.Greedy	Greedy sampling.
gm.text.RandomSampling	Simple random sampling.
gm.text.Sampler	Sampler.
gm.text.SamplingMethod	Base class for sampling methods.
gm.text.SpecialTokens	Special tokens ids.
gm.text.Tokenizer	Base class for tokenizers.
gm.text.ToolSampler	Sampler with tool support.
gm.text.TopPSampling	Top-p (Nucleus) Sampling.
gm.text.TopkSampling	Top-k sampling.
gm.tools.McpToolHandler	Mcp tool handler.
gm.tools.ToolHandlerBase	Base class to orchestrate tools.

Function

gm.ckpts.load_params	Restore the params from a checkpoint.
gm.ckpts.save_params	Save the params to a checkpoint.
gm.data.make_seq2seq_fields	Create the model input, target and loss_mask.
gm.data.pad	Add zeros to the end of the sequence to reach the max length.
gm.math.apply_rope	Applies RoPE.
gm.math.count_consecutive	Counts consecutive identical elements in a list.
gm.testing.use_hermetic_tokenizer	Use the local tokenizer, to avoid TFHub calls.

Typing

gm.nn.Cache
gm.typing.Params