Feat: add Magistral Small 2509 and native mistral3 tokenizer support (#3165)

* feat: update mistral common

* feat: add mistral3processor

* fix: loading

* fix: cast pixel_values to fp32

* fix: image tensor conversion

* feat: add FA2 support for pixtral based models

* fix: update mistral small 3.1 to use native tokenizer

* fix: install tips

* fix: improve info on sample dataset files

* chore: move mistral configs into subfolders

* fix: remove unneeded patch

* fix: indent

* feat: add integration tests

* chore: move

* feat: add magistral 2509 docs and example

* fix: convert tensor to bool

* feat: expand tests

* chore: move tests

examples/mistral/mixtral/mixtral-8x22b-qlora-fsdp.yml

@@ -0,0 +1,76 @@
+base_model: mistral-community/Mixtral-8x22B-v0.1
+# optionally might have model_type or tokenizer_type
+model_type: AutoModelForCausalLM
+tokenizer_type: LlamaTokenizer
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+load_in_8bit: false
+load_in_4bit: true
+
+datasets:
+  - path: tatsu-lab/alpaca
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+output_dir: ./outputs/qlora-out
+
+model_config:
+  output_router_logits: true
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 1024
+sample_packing: false
+pad_to_sequence_len: false
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 1
+optimizer: adamw_torch_fused
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: true
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+loss_watchdog_threshold: 5.0
+loss_watchdog_patience: 3
+
+warmup_ratio: 0.1
+evals_per_epoch: 4
+saves_per_epoch: 1
+
+weight_decay: 0.0
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_limit_all_gathers: true
+  fsdp_sync_module_states: true
+  fsdp_offload_params: true
+  fsdp_use_orig_params: false
+  fsdp_cpu_ram_efficient_loading: true
+  fsdp_transformer_layer_cls_to_wrap: MixtralSparseMoeBlock
+  fsdp_state_dict_type: FULL_STATE_DICT
+  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
+special_tokens:
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/mistral/mixtral/mixtral-qlora-fsdp.yml

@@ -0,0 +1,81 @@
+base_model: mistralai/Mixtral-8x7B-v0.1
+# optionally might have model_type or tokenizer_type
+model_type: AutoModelForCausalLM
+tokenizer_type: LlamaTokenizer
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: true
+
+datasets:
+  - path: tatsu-lab/alpaca
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+output_dir: ./outputs/qlora-out
+
+model_config:
+  output_router_logits: true
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 1024
+sample_packing: false
+pad_to_sequence_len: false
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 4
+micro_batch_size: 2
+num_epochs: 1
+optimizer: adamw_torch_fused
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: true
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+loss_watchdog_threshold: 5.0
+loss_watchdog_patience: 3
+
+warmup_ratio: 0.1
+evals_per_epoch: 4
+saves_per_epoch: 1
+
+weight_decay: 0.0
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_limit_all_gathers: true
+  fsdp_sync_module_states: true
+  fsdp_offload_params: true
+  fsdp_use_orig_params: false
+  fsdp_cpu_ram_efficient_loading: true
+  fsdp_transformer_layer_cls_to_wrap: MixtralSparseMoeBlock
+  fsdp_state_dict_type: FULL_STATE_DICT
+  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
+  fsdp_sharding_strategy: FULL_SHARD
+  fsdp_forward_prefetch: false
+  fsdp_backward_prefetch: BACKWARD_PRE
+special_tokens:
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/mistral/mixtral/mixtral.yml

@@ -0,0 +1,85 @@
+base_model: mistralai/Mixtral-8x7B-v0.1
+# optionally might have model_type or tokenizer_type
+model_type: AutoModelForCausalLM
+tokenizer_type: LlamaTokenizer
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+trust_remote_code: true
+
+load_in_8bit: false
+load_in_4bit: true
+
+datasets:
+  - path: tatsu-lab/alpaca
+    type: alpaca
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.0
+output_dir: ./outputs/qlora-out
+
+## You can optionally freeze the entire model and unfreeze a subset of parameters
+unfrozen_parameters:
+#  - ^lm_head.weight$
+#  - ^model.embed_tokens.weight$[:32000]
+#  - model.layers.2[0-9]+.block_sparse_moe.gate
+#  - model.layers.2[0-9]+.block_sparse_moe.experts
+#  - model.layers.3[0-9]+.block_sparse_moe.gate
+#  - model.layers.3[0-9]+.block_sparse_moe.experts
+
+model_config:
+  output_router_logits: true
+
+adapter: qlora
+lora_model_dir:
+
+sequence_len: 4096
+sample_packing: true
+
+
+lora_r: 32
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_linear: true
+#lora_target_modules:
+#  - gate
+#  - q_proj
+#  - k_proj
+#  - v_proj
+#  - o_proj
+#  - w1
+#  - w2
+#  - w3
+
+wandb_project:
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+gradient_accumulation_steps: 2
+micro_batch_size: 1
+num_epochs: 1
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0002
+
+bf16: auto
+tf32: false
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+loss_watchdog_threshold: 5.0
+loss_watchdog_patience: 3
+
+warmup_ratio: 0.1
+evals_per_epoch: 4
+saves_per_epoch: 1
+
+deepspeed: deepspeed_configs/zero2.json
+weight_decay: 0.0
+special_tokens:
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

examples/mistral/mixtral/mixtral_22.yml

@@ -0,0 +1,55 @@
+base_model: mistral-community/Mixtral-8x22B-v0.1
+# optionally might have model_type or tokenizer_type
+model_type: AutoModelForCausalLM
+tokenizer_type: LlamaTokenizer
+# Automatically upload checkpoint and final model to HF
+# hub_model_id: username/custom_model_name
+
+trust_remote_code: true
+
+unfrozen_parameters:
+  - ^lm_head.weight$
+  - ^model.embed_tokens.weight$
+  - model.layers.4[4-9]+.block_sparse_moe.gate
+  - model.layers.4[4-9]+.block_sparse_moe.experts
+  - model.layers.5[0-5]+.block_sparse_moe.gate
+  - model.layers.5[0-5]+.block_sparse_moe.experts
+
+model_config:
+  output_router_logits: true
+
+datasets:
+  - path: yahma/alpaca-cleaned
+    type: alpaca
+output_dir: ./outputs/out
+
+sequence_len: 8000
+sample_packing: true
+
+
+gradient_accumulation_steps: 1
+micro_batch_size: 1
+num_epochs: 3
+optimizer: adamw_bnb_8bit
+lr_scheduler: cosine
+learning_rate: 0.0001
+
+bf16: auto
+tf32: false
+
+gradient_checkpointing: true
+resume_from_checkpoint:
+logging_steps: 1
+flash_attention: true
+
+save_total_limit: 1
+save_steps:
+
+deepspeed: deepspeed_configs/zero3_bf16_cpuoffload_all.json
+weight_decay: 0.0
+special_tokens:
+  eos_token: "<|im_end|>"
+tokens:
+  - "<|im_start|>"
+
+# save_first_step: true  # uncomment this to validate checkpoint saving works with your config