初始化项目，由ModelHub XC社区提供模型

Model: atrost/test_steerable_hf_model_v4
Source: Original Platform

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,199 @@
+---
+library_name: transformers
+tags: []
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]

chat_template.jinja

@@ -0,0 +1 @@
+{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜><think>\n'}}{% endif %}

config.json

@@ -0,0 +1,63 @@
+{
+  "architectures": [
+    "Qwen2ForCausalLMPostBlockSteeringFixed"
+  ],
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoModel": "qwen2_postblock_steering_fixed.Qwen2ModelPostBlockSteering",
+    "AutoModelForCausalLM": "qwen2_postblock_steering_fixed.Qwen2ForCausalLMPostBlockSteeringFixed"
+  },
+  "bos_token_id": 151643,
+  "dtype": "bfloat16",
+  "eos_token_id": 151643,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 8960,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 131072,
+  "max_window_layers": 21,
+  "model_type": "qwen2",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.57.3",
+  "use_cache": true,
+  "use_mrope": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,9 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 151646,
+  "do_sample": true,
+  "eos_token_id": 151643,
+  "temperature": 0.6,
+  "top_p": 0.95,
+  "transformers_version": "4.57.3"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a71d392648143015bc5b5d56961e08a6c83c679b8513ffec9ed28303e5f0d3d
+size 3555597480

qwen2_postblock_steering_fixed.py

@@ -0,0 +1,354 @@
+import torch
+import os
+import torch.nn as nn
+from typing import Optional, Tuple, Iterable, Union
+
+from transformers.models.qwen2.modeling_qwen2 import (
+    Qwen2ForCausalLM,
+    Qwen2Model,
+    Qwen2DecoderLayer,
+)
+
+# -------------------------
+# Low-rank adapter
+# -------------------------
+
+def _get_activation(name: str):
+    name = name.lower()
+    if name in ("silu", "swish"):
+        return nn.SiLU()
+    if name == "relu":
+        return nn.ReLU()
+    if name == "gelu":
+        return nn.GELU()
+    if name == "tanh":
+        return nn.Tanh()
+    raise ValueError(f"Unknown activation: {name}")
+
+class LowRankAdapter(nn.Module):
+    """
+    Δh = α * W_up( act(W_down(h)) )
+    """
+    def __init__(self, hidden_size: int, rank: int, alpha: float, activation: str):
+        super().__init__()
+        self.alpha = float(alpha)
+        self.act = _get_activation(activation)
+        self.down = nn.Linear(hidden_size, rank, bias=False)
+        self.up   = nn.Linear(rank, hidden_size, bias=False)
+
+        # start as no-op => preserves pretrained behavior at init
+        nn.init.zeros_(self.up.weight)
+
+    def forward(self, h: torch.Tensor) -> torch.Tensor:
+        return self.alpha * self.up(self.act(self.down(h)))
+
+
+# -------------------------
+# Steered Decoder Layer (post-block only)
+# -------------------------
+
+class Qwen2DecoderLayerPostBlockSteering(Qwen2DecoderLayer):
+    """
+    Drop-in Qwen2DecoderLayer that adds an adapter AFTER the block output.
+
+    apply_to:
+      - "last": apply only to last token (B,S,H) -> only position -1
+      - "all":  apply to all tokens
+    """
+    def __init__(
+        self,
+        config,
+        layer_idx: int,
+        # Custom arguments with defaults
+        enable: bool = True,
+        rank: int = 8,
+        alpha: float = 1.0,
+        activation: str = "silu",
+        apply_to: str = "all",
+        **kwargs  # <--- Best Practice: Catch any extra args the parent might need in future versions
+    ):
+        super().__init__(config, layer_idx, **kwargs)
+        assert apply_to in ("last", "all")
+        self.apply_to = apply_to
+        self._adapter_enabled = True
+
+        self.adapter_block = (
+            LowRankAdapter(
+                hidden_size=config.hidden_size,
+                rank=rank,
+                alpha=alpha,
+                activation=activation,
+            )
+            if enable
+            else None
+        )
+
+    def set_adapter_enabled(self, enabled: bool):
+        self._adapter_enabled = bool(enabled)
+
+    def _apply_last(self, x: torch.Tensor, adapter: nn.Module) -> torch.Tensor:
+        if x.ndim != 3:
+            return x
+        last = x[:, -1, :]  # (B,H)
+        new_last = (last + adapter(last)).unsqueeze(1)  # (B,1,H)
+        return torch.cat([x[:, :-1, :], new_last], dim=1)
+
+    def _apply_all(self, x: torch.Tensor, adapter: nn.Module) -> torch.Tensor:
+        if x.ndim != 3:
+            return x
+        b, s, h = x.shape
+        flat = x.reshape(b * s, h)
+        delta = adapter(flat).reshape(b, s, h)
+        return x + delta
+
+    def _apply_adapter(self, x: torch.Tensor) -> torch.Tensor:
+        if (self.adapter_block is None) or (not self._adapter_enabled):
+            return x
+        if self.apply_to == "last":
+            return self._apply_last(x, self.adapter_block)
+        return self._apply_all(x, self.adapter_block)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,  # legacy name
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: Optional[Tuple[torch.Tensor, torch.FloatTensor]] = None,
+        **kwargs,
+    ):
+        # Standard Qwen2 layer, inject adapter at the very end (post-block).
+
+        # NOTE: transformers 4.57+ Qwen2 expects decoder layers to return a Tensor
+        # (and optionally attn weights), NOT (hidden_states, ..., present_kv).
+        # Cache is carried via `past_key_values` (new API) and/or handled internally.
+
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Avoid passing BOTH past_key_value and past_key_values to attention.
+        past_key_values = kwargs.pop("past_key_values", None)
+        attn_kwargs = dict(kwargs)
+
+        if past_key_values is not None:
+            attn_kwargs["past_key_values"] = past_key_values
+            # do NOT also pass legacy past_key_value
+            pkv_arg = {}
+        else:
+            pkv_arg = {"past_key_value": past_key_value} if past_key_value is not None else {}
+
+        attn_out = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+            **pkv_arg,
+            **attn_kwargs,
+        )
+
+        # HF attention returns (attn_output,) or (attn_output, attn_weights)
+        if isinstance(attn_out, tuple):
+            attn_output = attn_out[0]
+            attn_weights = attn_out[1] if (output_attentions and len(attn_out) > 1) else None
+        else:
+            attn_output = attn_out
+            attn_weights = None
+
+        hidden_states = residual + attn_output
+
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = residual + self.mlp(hidden_states)
+
+        # ✅ post-block steering
+        hidden_states = self._apply_adapter(hidden_states)
+
+        # Return a Tensor (or Tensor + attn weights if requested). Do NOT return cache.
+        if output_attentions:
+            return (hidden_states, attn_weights)
+        return hidden_states
+
+
+
+# -------------------------
+# Qwen2Model + hardcoded steering config
+# -------------------------
+
+class Qwen2ModelPostBlockSteering(Qwen2Model):
+    def __init__(
+        self,
+        config,
+        layers_to_steer: Union[str, Iterable[int]] = "all",
+        rank: int = 8,
+        apply_to: str = "all",
+        alpha: float = 1.0,
+        activation: str = "silu",
+    ):
+        super().__init__(config)
+
+        if layers_to_steer == "all":
+            layer_ids = set(range(config.num_hidden_layers))
+        else:
+            layer_ids = set(int(i) for i in layers_to_steer)
+
+        new_layers = nn.ModuleList()
+        for i in range(config.num_hidden_layers):
+            new_layers.append(
+                Qwen2DecoderLayerPostBlockSteering(
+                    config=config,
+                    layer_idx=i,
+                    enable=(i in layer_ids),
+                    rank=rank,
+                    alpha=alpha,
+                    activation=activation,
+                    apply_to=apply_to,
+                )
+            )
+        self.layers = new_layers
+
+    def set_adapter_enabled(self, enabled: bool):
+        for layer in self.layers:
+            if hasattr(layer, "set_adapter_enabled"):
+                layer.set_adapter_enabled(enabled)
+
+
+# -------------------------
+# Qwen2ForCausalLM with hardcoded knobs + base frozen by default
+# -------------------------
+
+class Qwen2ForCausalLMPostBlockSteeringFixed(Qwen2ForCausalLM):
+    """
+    Hardcoded steering config + base frozen by default.
+
+    Change these class constants to match what you want globally.
+    """
+    STEER_RANK: int = 8
+    STEER_APPLY_TO: str = "last"         # "last" or "all"
+    STEER_LAYERS: Union[str, Iterable[int]] = "all"  # or e.g. [0, 5, 10]
+    STEER_ALPHA: float = 1.0
+    STEER_ACTIVATION: str = "silu"
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        # Replace base transformer with steered one using hardcoded config
+        self.model = Qwen2ModelPostBlockSteering(
+            config,
+            layers_to_steer=self.STEER_LAYERS,
+            rank=self.STEER_RANK,
+            apply_to=self.STEER_APPLY_TO,
+            alpha=self.STEER_ALPHA,
+            activation=self.STEER_ACTIVATION,
+        )
+
+        # Freeze base by default (only steering trainable)
+        self.freeze_base_keep_steering_trainable()
+
+    # ---- freezing / params ----
+
+    def freeze_base_keep_steering_trainable(self):
+        for n, p in self.named_parameters():
+            p.requires_grad = ("adapter_block" in n)
+
+    def steering_parameters(self):
+        for n, p in self.named_parameters():
+            if "adapter_block" in n:
+                yield p
+
+    # ---- dtype/device correctness for device_map="auto" ----
+
+    def cast_adapters_like_base(self):
+        """
+        If you load with torch_dtype="auto" and/or device_map="auto",
+        adapters are newly-created modules and need to match each layer’s dtype/device.
+        """
+        for layer in self.model.layers:
+            ref = layer.input_layernorm.weight
+            if getattr(layer, "adapter_block", None) is not None:
+                layer.adapter_block.to(device=ref.device, dtype=ref.dtype)
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        model = super().from_pretrained(*args, **kwargs)
+        # Ensure adapters are on the right shards/dtype, then freeze base
+        if hasattr(model, "cast_adapters_like_base"):
+            model.cast_adapters_like_base()
+        if hasattr(model, "freeze_base_keep_steering_trainable"):
+            model.freeze_base_keep_steering_trainable()
+        return model
+
+    def _prepare_for_serialization(self):
+        """
+        If the model was loaded with device_map/offload, Accelerate attaches hooks that
+        can break save_pretrained for newly-added params (like adapter_block.*).
+        This removes those hooks and consolidates to CPU.
+        """
+        try:
+            from accelerate.hooks import remove_hook_from_module
+            remove_hook_from_module(self, recurse=True)
+        except Exception:
+            pass
+
+        # Clean up common accelerate attributes if present
+        for attr in ("hf_device_map", "_hf_hook"):
+            if hasattr(self, attr):
+                try:
+                    delattr(self, attr)
+                except Exception:
+                    pass
+
+        # Ensure all params are materialized on CPU for a normal state_dict save
+        self.to("cpu")
+
+    def _strip_accelerate_offload_hooks(self):
+        """
+        Remove Accelerate's device_map/offload hooks so saving doesn't go through
+        get_state_dict_from_offload (which doesn't know about new adapter params).
+        """
+        # Best-effort official removers
+        try:
+            from accelerate.hooks import remove_hook_from_module
+            remove_hook_from_module(self, recurse=True)  # documented API :contentReference[oaicite:3]{index=3}
+        except Exception:
+            pass
+
+        # Hard removal: delete _hf_hook from every submodule if still present
+        for m in self.modules():
+            if hasattr(m, "_hf_hook"):
+                # try to detach cleanly if possible
+                try:
+                    m._hf_hook.detach_hook(m)
+                except Exception:
+                    pass
+                try:
+                    delattr(m, "_hf_hook")
+                except Exception:
+                    pass
+
+        # device_map bookkeeping (common on big-model inference)
+        if hasattr(self, "hf_device_map"):
+            try:
+                delattr(self, "hf_device_map")
+            except Exception:
+                pass
+
+    def save_pretrained(self, save_directory, **kwargs):
+        os.makedirs(save_directory, exist_ok=True)
+
+        # 1) remove accelerate offload hooks
+        self._strip_accelerate_offload_hooks()
+
+        # 2) consolidate to CPU (you cannot save sharded/offloaded weights “in place”)
+        self.to("cpu")
+
+        # 3) create a normal state_dict and pass it explicitly to bypass accelerate offload-saving
+        # (save_pretrained supports state_dict=...) :contentReference[oaicite:4]{index=4}
+        sd = {k: v.cpu() for k, v in self.state_dict().items()}
+
+        return super().save_pretrained(save_directory, state_dict=sd, **kwargs)

special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<｜begin▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e20ddafc659ba90242154b55275402edeca0715e5dbb30f56815a4ce081f4893
+size 11422778

tokenizer_config.json

@@ -0,0 +1,194 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<｜end▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<｜User｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151645": {
+      "content": "<｜Assistant｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151646": {
+      "content": "<｜begin▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|EOT|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151648": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151649": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "extra_special_tokens": {},
+  "legacy": true,
+  "model_max_length": 16384,
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizerFast",
+  "unk_token": null,
+  "use_default_system_prompt": false
+}