Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* 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 +1,189 @@
+---
+license: apache-2.0
+language:
+- en
+library_name: transformers
+tags:
+- code
+- python
+- maincoder
+- code-generation
+- reinforcement-learning
+- mcpo
+pipeline_tag: text-generation
+base_model: Maincode/Maincoder-1B
+---
+<img src="https://huggingface.co/datasets/Maincode/assets/resolve/e51154e034201be1a5dad0e9c8de31d8b9f17643/maincoder_logo.png" alt="" width="1250">
+
+[**Maincoder-1B**](https://maincode.com/maincoder/) is a code-focused language model optimized for code generation and completion tasks. The model achieves strong performance on coding benchmarks while maintaining a compact size suitable for local deployment.
+
+# Key Features
+
+- **Code Generation**: Optimized for Python code completion and generation tasks.
+- **Compact Size**: 1 billion parameters, lightweight enough to run on consumer hardware.
+- **Deep Architecture**: Modern transformer architecture with RoPE embeddings, grouped-query attention, QK normalization and high depth-to-width ratio.
+- **Advanced Data Mixing**: Pre-trained and mid-trained on custom data mixes developed for high-performance coding.
+- **MCPO Algorithm**: Fine-tuned with specialised reinforcement learning policy optimisation algorithm to improve training stability and accelerate convergence.
+- **SOTA Performance**: State-of-the-art performance on Python coding benchmarks HumanEval, HumanEval+ and MBPP+.
+
+# Benchmark Results
+
+<img src="https://huggingface.co/datasets/Maincode/assets/resolve/main/performance_h.png" alt="Benchmark Performance Across Baseline LLMs" width="1050">
+
+| Model | HumanEval | HumanEval+ | MBPP+ | MMLU | GSM8K |
+|---|---:|---:|---:|---:|---:|
+| [Maincode/Maincoder-1B](https://huggingface.co/Maincode/Maincoder-1B) | **0.7622** | **0.7256** |  **0.7090** | 0.3054 | 0.2976 |
+| [deepseek-ai/deepseek-coder-1.3b-instruct](https://huggingface.co/deepseek-ai/deepseek-coder-1.3b-instruct) | 0.5610 | 0.5305 |  0.6217 | 0.2705 | 0.0413 |
+| [HuggingFaceTB/SmolLM3-3B](https://huggingface.co/HuggingFaceTB/SmolLM3-3B) | 0.5366 | 0.5000 | 0.6799 | **0.5928** | 0.5505 |
+| [Qwen/Qwen2.5-Coder-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct) | 0.4634 | 0.4451 | 0.6561 | 0.4984 | 0.4944 |
+| [Qwen/Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B) |  0.4024 | 0.3780 | 0.5582 | 0.5571 |**0.6865** |
+
+# Model Overview
+
+Maincoder uses a modern transformer decoder architecture with:
+
+- **Rotary Position Embeddings**: With theta of 1,000,000.
+- **RMSNorm**: Pre-normalization for stable training.
+- **Grouped Query Attention**: 4:1 ratio of query to key-value heads.
+- **QK Normalization**: RMSNorm applied to attention queries and keys.
+- **SwiGLU MLP**: Gated linear units with SiLU activation.
+
+| Attribute | Value |
+|-----------|-------|
+| Parameters | 1B |
+| Hidden Size | 1536 |
+| Layers | 32 |
+| Attention Heads | 16 (4 KV heads) |
+| Head Dimension | 96 |
+| Vocabulary Size | 151,936 |
+| Context Length | 2,048 |
+| Precision | bfloat16 |
+
+# Usage
+
+### Installation
+
+```bash
+pip install transformers torch
+```
+
+### Quick Start
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+model = AutoModelForCausalLM.from_pretrained(
+    "Maincode/Maincoder-1B",
+    torch_dtype="auto",
+    device_map="auto",
+    trust_remote_code=True,
+)
+tokenizer = AutoTokenizer.from_pretrained(
+    "Maincode/Maincoder-1B",
+    trust_remote_code=True,
+)
+
+# Code completion example
+prompt = '''def fibonacci(n: int) -> int:
+    """Return the n-th Fibonacci number."""
+'''
+
+inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+outputs = model.generate(
+    **inputs,
+    max_new_tokens=256,
+    temperature=0.2,
+    do_sample=True,
+)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+```
+
+### Code Completion
+
+```python
+# Function completion
+prompt = '''def quicksort(arr: list) -> list:
+    """Sort a list using the quicksort algorithm."""
+'''
+
+# Class completion
+prompt = '''class BinarySearchTree:
+    """A binary search tree implementation."""
+    
+    def __init__(self):
+'''
+
+# Algorithm implementation
+prompt = '''def dijkstra(graph: dict, start: str, end: str) -> tuple:
+    """Find the shortest path using Dijkstra's algorithm.
+    
+    Args:
+        graph: Adjacency list representation of the graph
+        start: Starting node
+        end: Target node
+    
+    Returns:
+        Tuple of (distance, path)
+    """
+'''
+```
+
+# Additional Notes
+
+## Reproducibility
+
+<details>
+<summary>Model evaluations were run on 8 AMD MI355X GPUs via the <a href="https://github.com/EleutherAI/lm-evaluation-harness">EleutherAI</a> framework.</summary>
+
+```bash
+docker run --rm -it \
+  --device=/dev/kfd --device=/dev/dri --group-add=video \
+  --ipc=host --security-opt seccomp=unconfined \
+  -v $(pwd):/workspace -w /workspace \
+  -e HF_TOKEN \
+  -e PYTHONHASHSEED=0 \
+  -e TORCH_DETERMINISTIC=1 \
+  -e ROCBLAS_ATOMICS_MODE="0" \
+  -e MIOPEN_FIND_MODE="1" \
+  -e CUBLAS_WORKSPACE_CONFIG=":4096:8" \
+  -e HF_ALLOW_CODE_EVAL="1" \
+  rocm/pytorch:rocm7.1.1_ubuntu24.04_py3.12_pytorch_release_2.9.1 \
+  bash -c 'pip install "lm_eval[hf]" && \
+  accelerate launch -m lm_eval \
+  --model hf --model_args "pretrained=Maincode/Maincoder-1B,trust_remote_code=True,dtype=float32" \
+  --tasks humaneval,humaneval_plus,mbpp_plus,mmlu,gsm8k \
+  --device cuda:0 --batch_size 32 --seed 42 \
+  --confirm_run_unsafe_code'
+```
+
+</details>
+
+## Limitations
+
+- Context length limited to 2,048 tokens
+- Primarily optimized for Python, performance may vary on other languages
+- May generate code with bugs or security issues - always review generated code
+
+<div style="margin-left:14px; border-left:4px solid #3b82f6; background:rgba(59,130,246,0.08); padding:8px 10px; border-radius:8px; font-size:0.92em; margin:10px 0;">
+  <strong>Disclaimer</strong>: This model has <strong>not</strong> undergone any alignment or safety tuning (e.g., RLHF/RLAIF, DPO, or safety fine-tuning). Outputs may be unsafe or biased. Please use appropriate safeguards and evaluate carefully for your use case.
+</div>
+
+## License
+
+This model is released under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0).
+
+## Citation
+
+```bibtex
+@misc{maincoder2025,
+  title        = {Maincoder-1B: A High-Performance 1B Parameter Coding Model},
+  author       = {Maincode Team},
+  year         = {2025},
+  organization = {Maincode},
+  howpublished = {\url{https://huggingface.co/Maincode/Maincoder-1B}}
+}
+```
+
+## Contact
+
+For questions, issues, or collaboration inquiries, please visit [Maincode](https://maincode.com).

__init__.py

@@ -0,0 +1,30 @@
+# coding=utf-8
+# Copyright 2025 Maincode. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .configuration_maincoder import MaincoderConfig
+from .modelling_maincoder import (
+    MaincoderForCausalLM,
+    MaincoderModel,
+    MaincoderPreTrainedModel,
+)
+
+
+__all__ = [
+    "MaincoderConfig",
+    "MaincoderPreTrainedModel",
+    "MaincoderModel",
+    "MaincoderForCausalLM",
+]
+

added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

chat_template.jinja

@@ -0,0 +1,54 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- messages[0]['content'] }}
+    {%- else %}
+        {{- 'You are a helpful assistant.' }}
+    {%- endif %}
+    {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
+    {%- else %}
+        {{- '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- for message in messages %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role }}
+        {%- if message.content %}
+            {{- '\n' + message.content }}
+        {%- endif %}
+        {%- for tool_call in message.tool_calls %}
+            {%- if tool_call.function is defined %}
+                {%- set tool_call = tool_call.function %}
+            {%- endif %}
+            {{- '\n<tool_call>\n{"name": "' }}
+            {{- tool_call.name }}
+            {{- '", "arguments": ' }}
+            {{- tool_call.arguments | tojson }}
+            {{- '}\n</tool_call>' }}
+        {%- endfor %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

config.json

@@ -0,0 +1,34 @@
+{
+  "architectures": [
+    "MaincoderForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_maincoder.MaincoderConfig",
+    "AutoModel": "modelling_maincoder.MaincoderForCausalLM",
+    "AutoModelForCausalLM": "modelling_maincoder.MaincoderForCausalLM"
+  },
+  "bos_token_id": null,
+  "eos_token_id": 151643,
+  "head_dim": 96,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "intermediate_size_mlp": 4096,
+  "max_position_embeddings": 2048,
+  "model_type": "maincoder",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 4,
+  "pad_token_id": 151643,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.57.3",
+  "use_cache": true,
+  "use_qk_norm": true,
+  "vocab_size": 151936
+}

configuration_maincoder.py

@@ -0,0 +1,141 @@
+# coding=utf-8
+# Copyright 2025 Maincode. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Maincoder model configuration."""
+
+from typing import Optional
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+class MaincoderConfig(PretrainedConfig):
+    r"""
+    Configuration class for Maincoder model.
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 151936):
+            Vocabulary size of the Maincoder model.
+        hidden_size (`int`, *optional*, defaults to 1536):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 4096):
+            Dimension of the MLP intermediate representations.
+        intermediate_size_mlp (`int`, *optional*, defaults to 4096):
+            Dimension of the MLP representations (same as intermediate_size for dense models).
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 16):
+            Number of attention heads for each attention layer.
+        num_key_value_heads (`int`, *optional*, defaults to 4):
+            Number of key-value heads for Grouped Query Attention (GQA).
+        head_dim (`int`, *optional*, defaults to 96):
+            Dimension of each attention head.
+        hidden_act (`str`, *optional*, defaults to `"silu"`):
+            The activation function in the MLP.
+        max_position_embeddings (`int`, *optional*, defaults to 2048):
+            Maximum sequence length the model can handle.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            Standard deviation for weight initialization.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-05):
+            Epsilon for RMS normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether to use key-value cache for generation.
+        pad_token_id (`int`, *optional*, defaults to 151643):
+            Padding token id.
+        bos_token_id (`int`, *optional*):
+            Beginning of sequence token id.
+        eos_token_id (`int`, *optional*, defaults to 151643):
+            End of sequence token id.
+        tie_word_embeddings (`bool`, *optional*, defaults to `True`):
+            Whether to tie input and output embeddings.
+        rope_theta (`float`, *optional*, defaults to 1000000.0):
+            Base period for RoPE embeddings.
+        rope_scaling (`Dict`, *optional*):
+            RoPE scaling configuration for extended context.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            Dropout probability for attention weights.
+        use_qk_norm (`bool`, *optional*, defaults to `True`):
+            Whether to apply RMS normalization to query and key.
+
+    Example:
+    ```python
+    >>> from configuration_maincoder import MaincoderConfig
+    >>> from modelling_maincoder import MaincoderForCausalLM
+
+    >>> config = MaincoderConfig()
+    >>> model = MaincoderForCausalLM(config)
+    ```
+    """
+
+    model_type = "maincoder"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size: int = 151936,
+        hidden_size: int = 1536,
+        intermediate_size: int = 4096,
+        intermediate_size_mlp: int = 4096,
+        num_hidden_layers: int = 32,
+        num_attention_heads: int = 16,
+        num_key_value_heads: Optional[int] = 4,
+        head_dim: Optional[int] = 96,
+        hidden_act: str = "silu",
+        max_position_embeddings: int = 2048,
+        initializer_range: float = 0.02,
+        rms_norm_eps: float = 1e-5,
+        use_cache: bool = True,
+        pad_token_id: Optional[int] = 151643,
+        bos_token_id: Optional[int] = None,
+        eos_token_id: int = 151643,
+        tie_word_embeddings: bool = True,
+        rope_theta: float = 1000000.0,
+        rope_scaling: Optional[dict] = None,
+        attention_dropout: float = 0.0,
+        use_qk_norm: bool = True,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.intermediate_size_mlp = intermediate_size_mlp
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.max_position_embeddings = max_position_embeddings
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self.attention_dropout = attention_dropout
+        self.use_qk_norm = use_qk_norm
+        self.hidden_act = hidden_act
+
+        # GQA configuration
+        self.num_key_value_heads = num_key_value_heads if num_key_value_heads is not None else num_attention_heads
+        self.head_dim = head_dim if head_dim is not None else self.hidden_size // self.num_attention_heads
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+
+__all__ = ["MaincoderConfig"]

generation_config.json

@@ -0,0 +1,11 @@
+{
+  "_from_model_config": true,
+  "eos_token_id": [
+    151643,
+    128001,
+    128008,
+    128009
+  ],
+  "pad_token_id": 151643,
+  "transformers_version": "4.57.3"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fde6820b5be360f5ebc2d3136b3ced4d0c11976c58a4770fe61b8f11912cfac9
+size 2052447608

modelling_maincoder.py

@@ -0,0 +1,487 @@
+# coding=utf-8
+# Copyright 2025 Maincode. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Maincoder model implementation."""
+
+from typing import Callable, Optional, Union
+
+import torch
+import torch.nn as nn
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.generation import GenerationMixin
+from transformers.masking_utils import create_causal_mask
+from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
+from transformers.modeling_layers import GradientCheckpointingLayer
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.processing_utils import Unpack
+from transformers.utils import TransformersKwargs, auto_docstring, can_return_tuple, logging
+
+from .configuration_maincoder import MaincoderConfig
+
+
+logger = logging.get_logger(__name__)
+
+
+class MaincoderRMSNorm(nn.Module):
+    """RMSNorm implementation equivalent to T5LayerNorm."""
+
+    def __init__(self, hidden_size, eps=1e-5):
+        """
+        MatildaPlusRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.eps = eps
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+
+    def _norm(self, x):
+        return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
+
+    def forward(self, x):
+        output = self._norm(x.float()).type_as(x)
+        return output * self.weight
+
+    def extra_repr(self):
+        return f"{tuple(self.weight.shape)}, eps={self.eps}"
+
+
+class MaincoderMLP(nn.Module):
+    """SwiGLU-style MLP."""
+
+    def __init__(self, config: MaincoderConfig):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size_mlp
+
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+
+
+class MaincoderRotaryEmbedding(nn.Module):
+    """Rotary Position Embedding."""
+
+    def __init__(self, config: MaincoderConfig, device=None):
+        super().__init__()
+        self.rope_type = "llama3" if config.rope_scaling is not None else "default"
+        self.config = config
+        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+
+        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+    @torch.no_grad()
+    @dynamic_rope_update
+    def forward(self, x: torch.Tensor, position_ids: torch.Tensor) -> torch.Tensor:
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):
+            freqs = (inv_freq_expanded.to(x.device) @ position_ids_expanded).transpose(1, 2)
+            freqs_cis = torch.polar(torch.ones_like(freqs), freqs)
+            freqs_cis = freqs_cis * self.attention_scaling
+
+        return freqs_cis
+
+
+def apply_rotary_emb(
+    xq: torch.Tensor,
+    xk: torch.Tensor,
+    freqs_cis: torch.Tensor,
+) -> tuple[torch.Tensor, torch.Tensor]:
+    """Apply rotary embeddings to query and key tensors."""
+    xq_ = torch.view_as_complex(xq.float().reshape(*xq.shape[:-1], -1, 2))
+    xk_ = torch.view_as_complex(xk.float().reshape(*xk.shape[:-1], -1, 2))
+
+    # Broadcast freqs_cis
+    freqs_cis = freqs_cis[:, :, None, :]
+
+    xq_out = torch.view_as_real(xq_ * freqs_cis).flatten(3)
+    xk_out = torch.view_as_real(xk_ * freqs_cis).flatten(3)
+
+    return xq_out.type_as(xq), xk_out.type_as(xk)
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """Repeat key/value heads to match query heads for GQA."""
+    if n_rep == 1:
+        return hidden_states
+    batch, num_kv_heads, slen, head_dim = hidden_states.shape
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_kv_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_kv_heads * n_rep, slen, head_dim)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs,
+) -> tuple[torch.Tensor, torch.Tensor]:
+    """Eager attention implementation."""
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+
+    if attention_mask is not None:
+        attn_weights = attn_weights + attention_mask[:, :, :, : key_states.shape[-2]]
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+class MaincoderAttention(nn.Module):
+    """Multi-headed attention with Grouped Query Attention (GQA) and RoPE."""
+
+    def __init__(self, config: MaincoderConfig, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = config.head_dim
+        self.num_attention_heads = config.num_attention_heads
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_attention_heads // self.num_key_value_heads
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = config.attention_dropout
+
+        self.q_proj = nn.Linear(config.hidden_size, self.num_attention_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
+        self.o_proj = nn.Linear(self.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
+
+        # QK normalization
+        if config.use_qk_norm:
+            self.q_norm = MaincoderRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+            self.k_norm = MaincoderRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
+        batch_size, seq_len, _ = hidden_states.shape
+
+        query_states = self.q_proj(hidden_states).view(batch_size, seq_len, self.num_attention_heads, self.head_dim)
+        key_states = self.k_proj(hidden_states).view(batch_size, seq_len, self.num_key_value_heads, self.head_dim)
+        value_states = self.v_proj(hidden_states).view(batch_size, seq_len, self.num_key_value_heads, self.head_dim)
+
+        # Apply RoPE
+        query_states, key_states = apply_rotary_emb(query_states, key_states, position_embeddings)
+
+        # Apply QK normalization
+        if hasattr(self, "q_norm"):
+            query_states = self.q_norm(query_states)
+            key_states = self.k_norm(key_states)
+
+        # Transpose for attention: (batch, heads, seq, head_dim)
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        # Update KV cache
+        if past_key_values is not None:
+            cache_kwargs = {"cache_position": cache_position}
+            key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        # Attention
+        attention_fn: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_fn = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_fn(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(batch_size, seq_len, -1)
+        attn_output = self.o_proj(attn_output)
+
+        return attn_output, attn_weights
+
+
+class MaincoderDecoderLayer(GradientCheckpointingLayer):
+    """Transformer decoder layer with pre-norm architecture."""
+
+    def __init__(self, config: MaincoderConfig, layer_idx: int):
+        super().__init__()
+        self.self_attn = MaincoderAttention(config, layer_idx)
+        self.feed_forward = MaincoderMLP(config)
+        self.input_layernorm = MaincoderRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = MaincoderRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_embeddings: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> torch.Tensor:
+        # Self Attention
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            position_embeddings=position_embeddings,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            cache_position=cache_position,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        # Feed Forward
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.feed_forward(hidden_states)
+        hidden_states = residual + hidden_states
+
+        return hidden_states
+
+
+@auto_docstring
+class MaincoderPreTrainedModel(PreTrainedModel):
+    """Base class for Maincoder models."""
+
+    config_class = MaincoderConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MaincoderDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_sdpa = True
+    _supports_flex_attn = True
+
+    def _init_weights(self, module: nn.Module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+        elif isinstance(module, MaincoderRMSNorm):
+            module.weight.data.fill_(1.0)
+
+
+@auto_docstring
+class MaincoderModel(MaincoderPreTrainedModel):
+    """Maincoder transformer model outputting raw hidden states."""
+
+    def __init__(self, config: MaincoderConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList(
+            [MaincoderDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = MaincoderRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = MaincoderRotaryEmbedding(config)
+
+        self.post_init()
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> Union[tuple, BaseModelOutputWithPast]:
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache()
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = torch.arange(
+                past_seen_tokens,
+                past_seen_tokens + inputs_embeds.shape[1],
+                device=inputs_embeds.device,
+            )
+
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        # Create causal mask
+        causal_mask = create_causal_mask(
+            config=self.config,
+            input_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            past_key_values=past_key_values,
+        )
+
+        # Position embeddings
+        position_embeddings = self.rotary_emb(inputs_embeds, position_ids)
+
+        hidden_states = inputs_embeds
+        for layer in self.layers:
+            hidden_states = layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                position_embeddings=position_embeddings,
+                past_key_values=past_key_values,
+                cache_position=cache_position,
+                **kwargs,
+            )
+
+        hidden_states = self.norm(hidden_states)
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values if use_cache else None,
+        )
+
+
+class MaincoderForCausalLM(MaincoderPreTrainedModel, GenerationMixin):
+    """Maincoder model with a causal language modeling head."""
+
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config: MaincoderConfig):
+        super().__init__(config)
+        self.model = MaincoderModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        self.post_init()
+
+    def get_input_embeddings(self) -> nn.Embedding:
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value: nn.Embedding):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self) -> nn.Linear:
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings: nn.Linear):
+        self.lm_head = new_embeddings
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> Union[tuple, CausalLMOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer
+        >>> from modelling_maincoder import MaincoderForCausalLM
+
+        >>> model = MaincoderForCausalLM.from_pretrained("maincoder/maincoder")
+        >>> tokenizer = AutoTokenizer.from_pretrained("maincoder/maincoder")
+
+        >>> prompt = "def hello_world():"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=50)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True)[0]
+        ```"""
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+
+        # Only compute logits for tokens we need
+        if isinstance(logits_to_keep, int) and logits_to_keep > 0:
+            hidden_states = hidden_states[:, -logits_to_keep:, :]
+
+        logits = self.lm_head(hidden_states)
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs)
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+__all__ = [
+    "MaincoderConfig",
+    "MaincoderPreTrainedModel",
+    "MaincoderModel",
+    "MaincoderForCausalLM",
+]

special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c5ae00e602b8860cbd784ba82a8aa14e8feecec692e7076590d014d7b7fdafa
+size 11421896

tokenizer_config.json

@@ -0,0 +1,207 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "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
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 32768,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}