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

chat_template.jinja

@@ -0,0 +1,3 @@
+{% for message in messages %}{% if loop.first %}<|im_start|>{% endif %}{% if message['role'] == 'system' %}<|system|>{{ message['content'] }}<|endoftext|>
+{% endif %}{% if message['role'] == 'user' %}<|user|>{{ message['content'] }}<|endoftext|>
+<|assistant|>{% endif %}{% if message['role'] == 'assistant' %}{% generation %}{{ message['content'] }}<|endoftext|>{% endgeneration %}{% endif %}{% endfor %}

config.json

@@ -0,0 +1,44 @@
+{
+  "architectures": [
+    "NandiForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_nandi.NandiConfig",
+    "AutoModel": "modeling_nandi.NandiModel",
+    "AutoModelForCausalLM": "modeling_nandi.NandiForCausalLM"
+  },
+  "bos_token_id": 1,
+  "dtype": "bfloat16",
+  "embedding_rank": 768,
+  "eos_token_id": 0,
+  "factorized_embedding": false,
+  "head_dim": 78,
+  "hidden_act": "silu",
+  "hidden_size": 1248,
+  "initializer_range": 0.02,
+  "intermediate_size": 3556,
+  "kv_cache_mode": "shared",
+  "layer_sharing": false,
+  "layer_sharing_repeats": 2,
+  "max_position_embeddings": 2048,
+  "mlp_bias": false,
+  "model_type": "nandi",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 8,
+  "pad_token_id": 3,
+  "pretraining_tp": 1,
+  "qk_norm": true,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "rope_theta": 1000000.0,
+    "rope_type": "default"
+  },
+  "shared_kv": true,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.8.1",
+  "use_cache": false,
+  "vocab_size": 131072
+}

configuration_nandi.py

@@ -0,0 +1,119 @@
+from transformers.configuration_utils import PretrainedConfig
+
+
+class NandiConfig(PretrainedConfig):
+    r"""
+    Configuration class for the Nandi model.
+
+    Example:
+
+    ```python
+    >>> from transformers import AutoConfig, AutoModelForCausalLM
+
+    >>> configuration = AutoConfig.from_pretrained("Rta-AILabs/Nandi-500M-remote", trust_remote_code=True)
+
+    >>> model = AutoModelForCausalLM.from_pretrained("Rta-AILabs/Nandi-500M-remote", trust_remote_code=True)
+
+    >>> configuration = model.config
+    ```
+    """
+
+    model_type = "nandi"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    base_model_tp_plan = {
+        "layers.*.self_attn.q_proj": "colwise",
+        "layers.*.self_attn.k_proj": "colwise",
+        "layers.*.self_attn.v_proj": "colwise",
+        "layers.*.self_attn.o_proj": "rowwise",
+        "layers.*.mlp.gate_proj": "colwise",
+        "layers.*.mlp.up_proj": "colwise",
+        "layers.*.mlp.down_proj": "rowwise",
+    }
+
+    def __init__(
+        self,
+        vocab_size=131072,
+        hidden_size=1248,
+        intermediate_size=3556,
+        num_hidden_layers=28,
+        num_attention_heads=16,
+        num_key_value_heads=8,
+        head_dim=None,
+        hidden_act="silu",
+        max_position_embeddings=2048,
+        initializer_range=0.008,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=1,
+        eos_token_id=0,
+        pretraining_tp=1,
+        tie_word_embeddings=True,
+        rope_parameters=None,
+        attention_bias=False,
+        attention_dropout=0.0,
+        mlp_bias=False,
+        factorized_embedding=False,
+        embedding_rank=768,
+        layer_sharing=False,
+        layer_sharing_repeats=1,
+        qk_norm=True,
+        shared_kv=True,
+        kv_cache_mode="shared",
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        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 hidden_size // num_attention_heads
+        self.hidden_act = hidden_act
+        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.pretraining_tp = pretraining_tp
+        self.rope_parameters = rope_parameters if rope_parameters is not None else {"rope_theta": 1000000.0}
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+        self.mlp_bias = mlp_bias
+        self.factorized_embedding = factorized_embedding
+        self.embedding_rank = embedding_rank
+        self.layer_sharing = layer_sharing
+
+        self.layer_sharing_repeats = max(1, int(layer_sharing_repeats or 1))
+        self.qk_norm = qk_norm
+
+        self.shared_kv = shared_kv
+
+        if kv_cache_mode not in ("shared", "vanilla"):
+            raise ValueError(
+                f"`kv_cache_mode` must be 'shared' or 'vanilla', got {kv_cache_mode!r}."
+            )
+        self.kv_cache_mode = kv_cache_mode
+
+        if self.factorized_embedding and self.embedding_rank <= 0:
+            raise ValueError(
+                f"`embedding_rank` must be positive when `factorized_embedding=True`, got {self.embedding_rank}."
+            )
+        if self.hidden_size % self.num_attention_heads != 0:
+            raise ValueError(
+                f"`hidden_size` ({self.hidden_size}) must be divisible by "
+                f"`num_attention_heads` ({self.num_attention_heads})."
+            )
+        if self.layer_sharing_repeats < 1:
+            raise ValueError(f"`layer_sharing_repeats` must be >= 1, got {self.layer_sharing_repeats}.")
+
+        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__ = ["NandiConfig"]

generation_config.json

@@ -0,0 +1,10 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": [
+    0
+  ],
+  "pad_token_id": 3,
+  "transformers_version": "5.8.1",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de1486e789e46a5a654ac9d088440d35d6e80d6a2072c5096a54ef4ac55cf9c9
+size 1291104504

modeling_nandi.py

@@ -0,0 +1,533 @@
+from collections.abc import Callable
+
+import torch
+import torch.nn as nn
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache, DynamicLayer
+from transformers.generation import GenerationMixin
+from transformers.integrations import use_kernel_forward_from_hub
+from transformers.masking_utils import create_causal_mask
+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
+from transformers.utils.deprecation import deprecate_kwarg
+from transformers.utils.generic import can_return_tuple, merge_with_config_defaults
+from transformers.utils.output_capturing import capture_outputs
+from .configuration_nandi import NandiConfig
+
+
+@use_kernel_forward_from_hub("RMSNorm")
+class NandiRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+    def extra_repr(self):
+        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
+
+
+class NandiRotaryEmbedding(nn.Module):
+    inv_freq: torch.Tensor
+
+    def __init__(self, config: NandiConfig, device=None):
+        super().__init__()
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+        self.rope_type = self.config.rope_parameters.get("rope_type", "default")
+        rope_init_fn: Callable = self.compute_default_rope_parameters
+        if self.rope_type != "default":
+            rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+        inv_freq, self.attention_scaling = rope_init_fn(self.config, device)
+
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.register_buffer("original_inv_freq", inv_freq.clone(), persistent=False)
+
+    @staticmethod
+    def compute_default_rope_parameters(
+        config: NandiConfig | None = None,
+        device: torch.device | None = None,
+        seq_len: int | None = None,
+    ) -> tuple[torch.Tensor, float]:
+        del seq_len
+        base = config.rope_parameters["rope_theta"]
+        dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
+        attention_factor = 1.0
+        inv_freq = 1.0 / (
+            base ** (torch.arange(0, dim, 2, dtype=torch.int64).to(device=device, dtype=torch.float) / dim)
+        )
+        return inv_freq, attention_factor
+
+    @torch.no_grad()
+    @dynamic_rope_update
+    def forward(self, x, position_ids):
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
+        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.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+            cos = emb.cos() * self.attention_scaling
+            sin = emb.sin() * self.attention_scaling
+
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    del position_ids
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: torch.Tensor | None,
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs: Unpack[TransformersKwargs],
+):
+    del kwargs
+    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:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    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 NandiAttention(nn.Module):
+    def __init__(self, config: NandiConfig, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = config.head_dim
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = config.attention_dropout
+        self.is_causal = True
+
+        self.shared_kv = getattr(config, "shared_kv", False)
+
+        self.q_proj = nn.Linear(
+            config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.k_proj = nn.Linear(
+            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
+        )
+        if self.shared_kv:
+            self.v_proj = None
+        else:
+            self.v_proj = nn.Linear(
+                config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
+            )
+        self.o_proj = nn.Linear(
+            config.num_attention_heads * self.head_dim, config.hidden_size, bias=config.attention_bias
+        )
+
+        self.qk_norm = getattr(config, "qk_norm", False)
+        if self.qk_norm:
+            self.q_norm = NandiRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+            self.k_norm = NandiRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+        else:
+            self.q_norm = None
+            self.k_norm = None
+
+    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: torch.Tensor | None,
+        past_key_values: Cache | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        k_raw = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+
+        if self.shared_kv:
+            kv_cache_mode = getattr(self.config, "kv_cache_mode", "shared")
+
+            if self.qk_norm:
+                query_states = self.q_norm(query_states)
+
+            if kv_cache_mode == "shared":
+                if past_key_values is not None:
+                    empty_v = torch.empty(
+                        k_raw.shape[0],
+                        k_raw.shape[1],
+                        0,
+                        k_raw.shape[3],
+                        device=k_raw.device,
+                        dtype=k_raw.dtype,
+                    )
+                    k_raw_full, _ = past_key_values.update(k_raw, empty_v, self.layer_idx)
+                else:
+                    k_raw_full = k_raw
+
+                value_states = k_raw_full
+                key_states = self.k_norm(k_raw_full) if self.qk_norm else k_raw_full
+
+                cos, sin = position_embeddings
+                q_len = query_states.shape[-2]
+                cos_q = cos[..., -q_len:, :]
+                sin_q = sin[..., -q_len:, :]
+                query_states, _ = apply_rotary_pos_emb(query_states, query_states, cos_q, sin_q)
+                _, key_states = apply_rotary_pos_emb(key_states, key_states, cos, sin)
+
+            else:  
+                key_states = self.k_norm(k_raw) if self.qk_norm else k_raw
+                value_states = k_raw  
+
+                cos, sin = position_embeddings  
+                query_states, key_states = apply_rotary_pos_emb(
+                    query_states, key_states, cos, sin
+                )
+
+                if past_key_values is not None:
+                    key_states, value_states = past_key_values.update(
+                        key_states, value_states, self.layer_idx
+                    )
+
+        else:
+            value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+            key_states = k_raw
+
+            if self.qk_norm:
+                query_states = self.q_norm(query_states)
+                key_states = self.k_norm(key_states)
+
+            cos, sin = position_embeddings
+            query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+            if past_key_values is not None:
+                key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx)
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_interface(
+            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(*input_shape, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class NandiMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.gate_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=config.mlp_bias)
+        self.up_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=config.mlp_bias)
+        self.down_proj = nn.Linear(config.intermediate_size, config.hidden_size, bias=config.mlp_bias)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+
+
+class NandiDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: NandiConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = NandiAttention(config=config, layer_idx=layer_idx)
+        self.mlp = NandiMLP(config)
+        self.input_layernorm = NandiRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = NandiRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        use_cache: bool | None = False,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> torch.Tensor:
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+        return hidden_states
+
+
+class _VirtualLayerCache:
+    """Proxy that shifts cache layer indices by `offset` to give each repeat its own virtual slots."""
+
+    def __init__(self, cache: Cache, offset: int):
+        self._cache = cache
+        self._offset = offset
+
+    def __getattr__(self, name):
+        return getattr(self._cache, name)
+
+    def update(self, key_states, value_states, layer_idx, cache_kwargs=None):
+        virtual_idx = layer_idx + self._offset
+        # grow the backing cache if generate() pre-allocated fewer slots than needed
+        while len(self._cache.layers) <= virtual_idx:
+            self._cache.layers.append(DynamicLayer())
+        return self._cache.update(key_states, value_states, virtual_idx, cache_kwargs)
+
+    def get_seq_length(self, layer_idx: int = 0) -> int:
+        return self._cache.get_seq_length(layer_idx + self._offset)
+
+
+@auto_docstring
+class NandiPreTrainedModel(PreTrainedModel):
+    config: NandiConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["NandiDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+    _can_compile_fullgraph = True
+    _supports_attention_backend = True
+    _can_record_outputs = {
+        "hidden_states": NandiDecoderLayer,
+        "attentions": NandiAttention,
+    }
+
+    def __init__(self, config: NandiConfig):
+        super().__init__(config)
+
+
+@auto_docstring
+class NandiModel(NandiPreTrainedModel):
+    def __init__(self, config: NandiConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+        embedding_dim = config.embedding_rank if config.factorized_embedding else config.hidden_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, embedding_dim, self.padding_idx)
+        self.embedding_proj = (
+            nn.Linear(config.embedding_rank, config.hidden_size, bias=False) if config.factorized_embedding else None
+        )
+        self.layers = nn.ModuleList(
+            [NandiDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = NandiRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = NandiRotaryEmbedding(config=config)
+        self.gradient_checkpointing = False
+
+        self.post_init()
+
+    @merge_with_config_defaults
+    @capture_outputs
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        use_cache: bool | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> 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 self.embedding_proj is not None:
+            inputs_embeds = self.embedding_proj(inputs_embeds)
+        repeats = max(1, int(getattr(self.config, "layer_sharing_repeats", 1) or 1))
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache()
+
+        if position_ids is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            position_ids = torch.arange(inputs_embeds.shape[1], device=inputs_embeds.device) + past_seen_tokens
+            position_ids = position_ids.unsqueeze(0)
+
+        causal_mask = create_causal_mask(
+            config=self.config,
+            inputs_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+
+        hidden_states = inputs_embeds
+        kv_cache_mode = getattr(self.config, "kv_cache_mode", "shared")
+        if (
+            getattr(self.config, "shared_kv", False)
+            and kv_cache_mode == "shared"
+            and past_key_values is not None
+        ):
+            past_len = past_key_values.get_seq_length(0)
+            cur_len = inputs_embeds.shape[1]
+            full_position_ids = torch.arange(
+                past_len + cur_len, device=inputs_embeds.device
+            ).unsqueeze(0)
+            position_embeddings = self.rotary_emb(hidden_states, position_ids=full_position_ids)
+        else:
+            position_embeddings = self.rotary_emb(hidden_states, position_ids=position_ids)
+
+        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
+            for repeat_idx in range(repeats):
+
+                repeat_cache = (
+                    _VirtualLayerCache(past_key_values, repeat_idx * self.config.num_hidden_layers)
+                    if (past_key_values is not None and repeat_idx > 0)
+                    else past_key_values
+                )
+                hidden_states = decoder_layer(
+                    hidden_states,
+                    attention_mask=causal_mask,
+                    position_embeddings=position_embeddings,
+                    position_ids=position_ids,
+                    past_key_values=repeat_cache,
+                    use_cache=use_cache,
+                    **kwargs,
+                )
+
+        hidden_states = self.norm(hidden_states)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+        )
+
+
+@auto_docstring
+class NandiForCausalLM(NandiPreTrainedModel, GenerationMixin):
+    _tied_weights_keys = {"lm_head.weight": "model.embed_tokens.weight"}
+    _tp_plan = {"lm_head": "colwise_gather_output"}
+    _pp_plan = {
+        "lm_head_proj": (["hidden_states"], ["hidden_states"]),
+        "lm_head": (["hidden_states"], ["logits"]),
+    }
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = NandiModel(config)
+        self.vocab_size = config.vocab_size
+
+        lm_head_in_features = config.embedding_rank if config.factorized_embedding else config.hidden_size
+        self.lm_head_proj = (
+            nn.Linear(config.hidden_size, config.embedding_rank, bias=False) if config.factorized_embedding else None
+        )
+        self.lm_head = nn.Linear(lm_head_in_features, config.vocab_size, bias=False)
+
+        self.post_init()
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
+        use_cache: bool | None = None,
+        logits_to_keep: int | torch.Tensor = 0,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> CausalLMOutputWithPast:
+        outputs: BaseModelOutputWithPast = 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,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        if self.lm_head_proj is not None:
+            hidden_states = self.lm_head_proj(hidden_states)
+
+        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        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__ = ["NandiPreTrainedModel", "NandiModel", "NandiForCausalLM"]

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f9fd2911e5e02cb959f6a77a1ebd4bba088d4ec2e0bc0a208b3c1e0ca2278791
+size 12460626

tokenizer_config.json

@@ -0,0 +1,9 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|im_start|>",
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<|pad|>",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": "<|endoftext|>"
+}