| awaitSampledToken() | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inlineprotected |
| Base typedef (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | |
| componentPaths() const | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inline |
| enqueueSampleNext(const TensorType &logits, TokenTensor &token_out, const SamplingParams ¶ms) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inlineprotected |
| eosToken() const noexcept override | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inlineprotectedvirtual |
| finalLogitSoftcap() const noexcept | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inlineprotectedvirtual |
| fp4_group_size_ | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | protected |
| fromCheckpoint(const std::filesystem::path &path, DeviceId device_id=DeviceId{ TDeviceType, 0 }, dim_t context_length=0) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inlinestatic |
| generate(std::span< const int32_t > prompt_tokens, const std::function< void(int32_t)> &on_token, const GenerateParams ¶ms={}, std::stop_token stop={}) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inline |
| getConfig() const noexcept (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inline |
| getDeviceId() const noexcept | Mila::Dnn::Model< TDeviceType, TPrecision > | inline |
| getMemoryStats() const | Mila::Dnn::Model< TDeviceType, TPrecision > | inline |
| getNetwork() noexcept (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inlineprotected |
| getNetwork() const noexcept (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inlineprotected |
| getRuntimeMode() const noexcept | Mila::Dnn::Model< TDeviceType, TPrecision > | inline |
| GptModel(const GptModel &)=delete (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| GptModel(GptModel &&)=default (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| GptTransformerType typedef (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| isEval() const noexcept | Mila::Dnn::Model< TDeviceType, TPrecision > | inline |
| isInferenceMode() const noexcept | Mila::Dnn::Model< TDeviceType, TPrecision > | inline |
| isTrainingMode() const noexcept | Mila::Dnn::Model< TDeviceType, TPrecision > | inline |
| LanguageModel(const LanguageModel &)=delete (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | |
| LanguageModel(LanguageModel &&)=default (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | |
| LanguageModel(std::unique_ptr< LanguageModelNetwork< TDeviceType, TPrecision > > network, RuntimeMode runtime_mode, Serialization::WeightsMetadata source_metadata={}, WeightQuantization weight_quantization=WeightQuantization::None, int fp4_group_size=128) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inlineexplicitprotected |
| load(const std::filesystem::path &path, dim_t context_length, DeviceId device_id=DeviceId{ TDeviceType, 0 }, bool strict=true) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inlinestatic |
| maxSequenceLength() const noexcept override | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inlineprotectedvirtual |
| Model(const Model &)=delete (defined in Mila::Dnn::Model< TDeviceType, TPrecision >) | Mila::Dnn::Model< TDeviceType, TPrecision > | |
| Model(Model &&)=default (defined in Mila::Dnn::Model< TDeviceType, TPrecision >) | Mila::Dnn::Model< TDeviceType, TPrecision > | |
| Model(std::unique_ptr< NetworkType > network, RuntimeMode runtime_mode) | Mila::Dnn::Model< TDeviceType, TPrecision > | inlineexplicitprotected |
| ModelBase typedef (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| MR typedef (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| network_ | Mila::Dnn::Model< TDeviceType, TPrecision > | protected |
| NetworkType typedef (defined in Mila::Dnn::Model< TDeviceType, TPrecision >) | Mila::Dnn::Model< TDeviceType, TPrecision > | |
| observe(std::string_view pattern, Compute::ComputePassMask passes, Compute::ActivationObserver sink) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inline |
| onGenerating(std::span< const int32_t > prompt_tokens, const std::function< void(int32_t)> &on_token, const GenerateParams ¶ms, std::stop_token stop) override | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inlineprotectedvirtual |
| onTraining() override | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inlineprotectedvirtual |
| operator=(const GptModel &)=delete (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| operator=(GptModel &&)=default (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| operator=(const LanguageModel &)=delete (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | |
| operator=(LanguageModel &&)=default (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | |
| operator=(const Model &)=delete (defined in Mila::Dnn::Model< TDeviceType, TPrecision >) | Mila::Dnn::Model< TDeviceType, TPrecision > | |
| operator=(Model &&)=default (defined in Mila::Dnn::Model< TDeviceType, TPrecision >) | Mila::Dnn::Model< TDeviceType, TPrecision > | |
| sampleNext(const TensorType &logits, TokenTensor &token_out, const SamplingParams ¶ms) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inlineprotected |
| save(const std::filesystem::path &path) const | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inline |
| saveCheckpoint(const std::filesystem::path &path, SerializationMode mode=SerializationMode::Checkpoint) const | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inline |
| seedSampler(uint64_t seed) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inline |
| setEval(bool eval) | Mila::Dnn::Model< TDeviceType, TPrecision > | inline |
| source_metadata_ | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | protected |
| stopObserving() | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inline |
| stopTokens() const (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inlineprotectedvirtual |
| TensorType typedef (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| TokenIndexType typedef (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| TokenTensor typedef (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | protected |
| toString() const override | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inlinevirtual |
| train() | Mila::Dnn::Model< TDeviceType, TPrecision > | inline |
| vocabSize() const noexcept override | Mila::Dnn::GptModel< TDeviceType, TPrecision > | inlineprotectedvirtual |
| weight_quantization_ | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | protected |
| weightQuantizationScheme() const | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | inline |
| ~GptModel()=default (defined in Mila::Dnn::GptModel< TDeviceType, TPrecision >) | Mila::Dnn::GptModel< TDeviceType, TPrecision > | |
| ~LanguageModel()=default (defined in Mila::Dnn::LanguageModel< TDeviceType, TPrecision >) | Mila::Dnn::LanguageModel< TDeviceType, TPrecision > | virtual |
| ~Model()=default (defined in Mila::Dnn::Model< TDeviceType, TPrecision >) | Mila::Dnn::Model< TDeviceType, TPrecision > | virtual |
1.17.0