[Model] Support Step1 Model (#32511)

Signed-off-by: xieli <xieli@stepfun.com>

docs/models/supported_models.md

@@ -452,9 +452,10 @@ th {
 | `Qwen3MoeForCausalLM` | Qwen3MoE | `Qwen/Qwen3-30B-A3B`, etc. | ✅︎ | ✅︎ |
 | `Qwen3NextForCausalLM` | Qwen3NextMoE | `Qwen/Qwen3-Next-80B-A3B-Instruct`, etc. | ✅︎ | ✅︎ |
 | `SeedOssForCausalLM` | SeedOss | `ByteDance-Seed/Seed-OSS-36B-Instruct`, etc. | ✅︎ | ✅︎ |
+| `SolarForCausalLM` | Solar Pro | `upstage/solar-pro-preview-instruct`, etc. | ✅︎ | ✅︎ |
 | `StableLmForCausalLM` | StableLM | `stabilityai/stablelm-3b-4e1t`, `stabilityai/stablelm-base-alpha-7b-v2`, etc. | | |
 | `Starcoder2ForCausalLM` | Starcoder2 | `bigcode/starcoder2-3b`, `bigcode/starcoder2-7b`, `bigcode/starcoder2-15b`, etc. | | ✅︎ |
-| `SolarForCausalLM` | Solar Pro | `upstage/solar-pro-preview-instruct`, etc. | ✅︎ | ✅︎ |
+| `Step1ForCausalLM` | Step-Audio | `stepfun-ai/Step-Audio-EditX`, etc. | ✅︎ | ✅︎ |
 | `TeleChat2ForCausalLM` | TeleChat2 | `Tele-AI/TeleChat2-3B`, `Tele-AI/TeleChat2-7B`, `Tele-AI/TeleChat2-35B`, etc. | ✅︎ | ✅︎ |
 | `TeleFLMForCausalLM` | TeleFLM | `CofeAI/FLM-2-52B-Instruct-2407`, `CofeAI/Tele-FLM`, etc. | ✅︎ | ✅︎ |
 | `XverseForCausalLM` | XVERSE | `xverse/XVERSE-7B-Chat`, `xverse/XVERSE-13B-Chat`, `xverse/XVERSE-65B-Chat`, etc. | ✅︎ | ✅︎ |

tests/models/registry.py

@@ -472,6 +472,9 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
         "ByteDance-Seed/Seed-OSS-36B-Instruct",
         trust_remote_code=True,
     ),
+    "Step1ForCausalLM": _HfExamplesInfo(
+        "stepfun-ai/Step-Audio-EditX", trust_remote_code=True
+    ),
     "SmolLM3ForCausalLM": _HfExamplesInfo("HuggingFaceTB/SmolLM3-3B"),
     "StableLMEpochForCausalLM": _HfExamplesInfo("stabilityai/stablelm-zephyr-3b"),
     "StableLmForCausalLM": _HfExamplesInfo("stabilityai/stablelm-3b-4e1t"),

tests/models/test_initialization.py

@@ -115,8 +115,11 @@ def can_initialize(
         # FIXME: A hack to bypass FA3 assertion because our CI's L4 GPU
         # has cc==8.9 which hasn't supported FA3 yet. Remove this hack when
         # L4 supports FA3.
+        # Step1ForCausalLM requires TRITON_ATTN for use_alibi_sqrt support.
         attention_config = (
-            {"backend": "TRITON_ATTN"} if model_arch == "GptOssForCausalLM" else None
+            {"backend": "TRITON_ATTN"}
+            if model_arch in ("GptOssForCausalLM", "Step1ForCausalLM")
+            else None
         )
         if model_arch == "WhisperForConditionalGeneration":
             m.setenv("VLLM_WORKER_MULTIPROC_METHOD", "spawn")

vllm/attention/layer.py

@@ -162,6 +162,7 @@ class Attention(nn.Module, AttentionLayerBase):
         scale: float,
         num_kv_heads: int | None = None,
         alibi_slopes: list[float] | None = None,
+        use_alibi_sqrt: bool | None = None,
         cache_config: CacheConfig | None = None,
         quant_config: QuantizationConfig | None = None,
         logits_soft_cap: float | None = None,
@@ -243,7 +244,16 @@ class Attention(nn.Module, AttentionLayerBase):
             )
         else:
             self.attn_backend = attn_backend
-
+        backend_supports_alibi_sqrt = self.attn_backend.supports_alibi_sqrt()
+        use_alibi_sqrt = use_alibi_sqrt if use_alibi_sqrt else False
+        if use_alibi_sqrt and not backend_supports_alibi_sqrt:
+            raise ValueError(
+                f"use_alibi_sqrt is not supported by backend "
+                f"{self.attn_backend.get_name()}."
+            )
+        self.use_alibi_sqrt = bool(use_alibi_sqrt)
+        if backend_supports_alibi_sqrt:
+            extra_impl_args["use_alibi_sqrt"] = self.use_alibi_sqrt
         # prefix caching + batch invariance is currently not supported for
         # FLASHINFER and TRITON_MLA.
         if (

vllm/model_executor/models/registry.py

@@ -185,6 +185,7 @@ _TEXT_GENERATION_MODELS = {
     "Qwen3MoeForCausalLM": ("qwen3_moe", "Qwen3MoeForCausalLM"),
     "RWForCausalLM": ("falcon", "FalconForCausalLM"),
     "SeedOssForCausalLM": ("seed_oss", "SeedOssForCausalLM"),
+    "Step1ForCausalLM": ("step1", "Step1ForCausalLM"),
     "Step3TextForCausalLM": ("step3_text", "Step3TextForCausalLM"),
     "StableLMEpochForCausalLM": ("stablelm", "StablelmForCausalLM"),
     "StableLmForCausalLM": ("stablelm", "StablelmForCausalLM"),

vllm/model_executor/models/step1.py

+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""Shared Step decoder blocks and the Step1 text model."""
+
+from __future__ import annotations
+
+import math
+from collections.abc import Iterable
+
+import torch
+from torch import nn
+
+from vllm.attention.layer import Attention, AttentionType
+from vllm.config import CacheConfig, VllmConfig
+from vllm.distributed import (
+    get_pp_group,
+    get_tensor_model_parallel_rank,
+    get_tensor_model_parallel_world_size,
+)
+from vllm.model_executor.layers.activation import SiluAndMul
+from vllm.model_executor.layers.layernorm import RMSNorm
+from vllm.model_executor.layers.linear import (
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
+from vllm.model_executor.layers.logits_processor import LogitsProcessor
+from vllm.model_executor.layers.quantization import QuantizationConfig
+from vllm.model_executor.layers.vocab_parallel_embedding import (
+    ParallelLMHead,
+    VocabParallelEmbedding,
+)
+from vllm.model_executor.model_loader.weight_utils import default_weight_loader
+from vllm.model_executor.models.interfaces import SupportsPP
+from vllm.model_executor.models.utils import (
+    AutoWeightsLoader,
+    PPMissingLayer,
+    is_pp_missing_parameter,
+    make_empty_intermediate_tensors_factory,
+    make_layers,
+    maybe_prefix,
+)
+from vllm.sequence import IntermediateTensors
+
+STEP_PACKED_MODULES_MAPPING = {
+    "qkv_proj": ["q_proj", "k_proj", "v_proj"],
+    "gate_up_proj": ["gate_proj", "up_proj"],
+}
+
+
+def _get_step_alibi_slopes(total_num_heads: int) -> torch.Tensor:
+    """Reference ALiBi slopes used by Step models."""
+    closest_power_of_2 = 2 ** math.floor(math.log2(total_num_heads))
+    base = torch.tensor(
+        2 ** (-8.0 / closest_power_of_2),
+        dtype=torch.float32,
+    )
+    slopes = torch.pow(
+        base,
+        torch.arange(1, 1 + closest_power_of_2, dtype=torch.int32),
+    )
+    if closest_power_of_2 != total_num_heads:
+        extra_base = torch.tensor(
+            2 ** (-4.0 / closest_power_of_2),
+            dtype=torch.float32,
+        )
+        num_remaining_heads = total_num_heads - closest_power_of_2
+        extra_powers = torch.arange(
+            1,
+            1 + 2 * num_remaining_heads,
+            2,
+            dtype=torch.int32,
+        )
+        slopes = torch.cat(
+            [slopes, torch.pow(extra_base, extra_powers)],
+            dim=0,
+        )
+    return slopes
+
+
+class StepAttention(nn.Module):
+    def __init__(
+        self,
+        config,
+        cache_config: CacheConfig | None = None,
+        quant_config: QuantizationConfig | None = None,
+        prefix: str = "",
+    ):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        tp_size = get_tensor_model_parallel_world_size()
+        self.total_num_heads = config.num_attention_heads
+        assert self.total_num_heads % tp_size == 0
+        self.num_heads = self.total_num_heads // tp_size
+        self.head_dim = self.hidden_size // self.total_num_heads
+
+        total_num_kv_heads = getattr(
+            config, "num_attention_groups", getattr(config, "num_key_value_heads", 1)
+        )
+        if total_num_kv_heads is None or total_num_kv_heads <= 0:
+            total_num_kv_heads = 1
+        self.total_num_kv_heads = total_num_kv_heads
+        if self.total_num_kv_heads >= tp_size:
+            assert self.total_num_kv_heads % tp_size == 0
+        else:
+            assert tp_size % self.total_num_kv_heads == 0
+        self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size)
+
+        self.qkv_proj = QKVParallelLinear(
+            hidden_size=self.hidden_size,
+            head_size=self.head_dim,
+            total_num_heads=self.total_num_heads,
+            total_num_kv_heads=self.total_num_kv_heads,
+            bias=getattr(config, "attention_bias", False),
+            quant_config=quant_config,
+            prefix=f"{prefix}.qkv_proj",
+        )
+        self.q_size = self.num_heads * self.head_dim
+        self.kv_size = self.num_kv_heads * self.head_dim
+
+        self.o_proj = RowParallelLinear(
+            input_size=self.total_num_heads * self.head_dim,
+            output_size=self.hidden_size,
+            bias=getattr(config, "attention_bias", False),
+            quant_config=quant_config,
+            prefix=f"{prefix}.o_proj",
+        )
+
+        tp_rank = get_tensor_model_parallel_rank()
+        head_start = tp_rank * self.num_heads
+        head_end = (tp_rank + 1) * self.num_heads
+        alibi_slopes = _get_step_alibi_slopes(self.total_num_heads)[head_start:head_end]
+        alibi_slopes = alibi_slopes.tolist()
+
+        self.scale = self.head_dim**-0.5
+        self.attn = Attention(
+            self.num_heads,
+            self.head_dim,
+            self.scale,
+            num_kv_heads=self.num_kv_heads,
+            cache_config=cache_config,
+            quant_config=quant_config,
+            alibi_slopes=alibi_slopes,
+            prefix=f"{prefix}.attn",
+            use_alibi_sqrt=True,
+            attn_type=AttentionType.DECODER,
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+    ) -> torch.Tensor:
+        qkv, _ = self.qkv_proj(hidden_states)
+        q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
+        attn_output = self.attn(q, k, v)
+        output, _ = self.o_proj(attn_output)
+        return output
+
+
+class StepMLP(nn.Module):
+    def __init__(
+        self,
+        hidden_size: int,
+        intermediate_size: int,
+        quant_config: QuantizationConfig | None = None,
+        prefix: str = "",
+        bias: bool = False,
+    ):
+        super().__init__()
+        self.gate_up_proj = MergedColumnParallelLinear(
+            input_size=hidden_size,
+            output_sizes=[intermediate_size, intermediate_size],
+            bias=bias,
+            quant_config=quant_config,
+            prefix=f"{prefix}.gate_up_proj",
+        )
+        self.down_proj = RowParallelLinear(
+            input_size=intermediate_size,
+            output_size=hidden_size,
+            bias=bias,
+            quant_config=quant_config,
+            prefix=f"{prefix}.down_proj",
+        )
+        self.act_fn = SiluAndMul()
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x, _ = self.gate_up_proj(x)
+        x = self.act_fn(x)
+        x, _ = self.down_proj(x)
+        return x
+
+
+class StepDecoderLayer(nn.Module):
+    def __init__(self, vllm_config: VllmConfig, prefix: str = ""):
+        super().__init__()
+        config = vllm_config.model_config.hf_config
+        cache_config = vllm_config.cache_config
+        quant_config = vllm_config.quant_config
+
+        self.hidden_size = config.hidden_size
+        self.self_attn = StepAttention(
+            config=config,
+            cache_config=cache_config,
+            quant_config=quant_config,
+            prefix=f"{prefix}.self_attn",
+        )
+        self.mlp = StepMLP(
+            hidden_size=self.hidden_size,
+            intermediate_size=config.intermediate_size,
+            quant_config=quant_config,
+            prefix=f"{prefix}.mlp",
+            bias=getattr(config, "mlp_bias", False),
+        )
+        self.input_layernorm = RMSNorm(
+            self.hidden_size,
+            eps=config.rms_norm_eps,
+        )
+        self.post_attention_layernorm = RMSNorm(
+            self.hidden_size,
+            eps=config.rms_norm_eps,
+        )
+
+    def forward(
+        self,
+        positions: torch.Tensor,
+        hidden_states: torch.Tensor,
+        residual: torch.Tensor | None,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        if residual is None:
+            residual = hidden_states
+            hidden_states = self.input_layernorm(hidden_states)
+        else:
+            hidden_states, residual = self.input_layernorm(hidden_states, residual)
+        hidden_states = self.self_attn(hidden_states=hidden_states)
+        hidden_states, residual = self.post_attention_layernorm(hidden_states, residual)
+        hidden_states = self.mlp(hidden_states)
+        return hidden_states, residual
+
+    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
+        stacked_params_mapping = [
+            (".qkv_proj", ".q_proj", "q"),
+            (".qkv_proj", ".k_proj", "k"),
+            (".qkv_proj", ".v_proj", "v"),
+            (".gate_up_proj", ".gate_proj", 0),
+            (".gate_up_proj", ".up_proj", 1),
+        ]
+        params_dict = dict(self.named_parameters())
+        loaded_params: set[str] = set()
+        for name, loaded_weight in weights:
+            for param_name, weight_name, shard_id in stacked_params_mapping:
+                if weight_name not in name:
+                    continue
+                name = name.replace(weight_name, param_name)
+                # Skip loading extra bias for GPTQ models.
+                if name.endswith(".bias") and name not in params_dict:
+                    continue
+                if is_pp_missing_parameter(name, self):
+                    continue
+                param = params_dict[name]
+                weight_loader = param.weight_loader
+                weight_loader(param, loaded_weight, shard_id)
+                break
+            else:
+                # Skip loading extra bias for GPTQ models.
+                if name.endswith(".bias") and name not in params_dict:
+                    continue
+                if is_pp_missing_parameter(name, self):
+                    continue
+                param = params_dict[name]
+                weight_loader = getattr(param, "weight_loader", default_weight_loader)  # type: ignore[name-defined]
+                weight_loader(param, loaded_weight)
+            loaded_params.add(name)
+        return loaded_params
+
+
+class StepDecoderModel(nn.Module):
+    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
+        super().__init__()
+        config = vllm_config.model_config.hf_config
+        quant_config = vllm_config.quant_config
+
+        self.config = config
+        self.quant_config = quant_config
+        # Need embed_tokens on first rank, and also on last rank if tie_word_embeddings
+        if get_pp_group().is_first_rank or (
+            config.tie_word_embeddings and get_pp_group().is_last_rank
+        ):
+            self.embed_tokens = VocabParallelEmbedding(
+                config.vocab_size,
+                config.hidden_size,
+                quant_config=quant_config,
+            )
+        else:
+            self.embed_tokens = PPMissingLayer()
+        self.start_layer, self.end_layer, self.layers = make_layers(
+            config.num_hidden_layers,
+            lambda prefix: StepDecoderLayer(vllm_config=vllm_config, prefix=prefix),
+            prefix=maybe_prefix(prefix, "layers"),
+        )
+        if get_pp_group().is_last_rank:
+            self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        else:
+            self.norm = PPMissingLayer()
+
+        self.aux_hidden_state_layers: tuple[int, ...] = getattr(
+            config, "aux_hidden_state_layers", ()
+        )
+        self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
+            ["hidden_states", "residual"],
+            config.hidden_size,
+        )
+
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.embed_tokens(input_ids)
+
+    def forward(
+        self,
+        input_ids: torch.Tensor | None,
+        positions: torch.Tensor,
+        intermediate_tensors: IntermediateTensors | None,
+        inputs_embeds: torch.Tensor | None = None,
+    ) -> torch.Tensor | IntermediateTensors | tuple[torch.Tensor, list[torch.Tensor]]:
+        if get_pp_group().is_first_rank:
+            if inputs_embeds is not None:
+                hidden_states = inputs_embeds
+            else:
+                assert input_ids is not None
+                hidden_states = self.embed_input_ids(input_ids)
+            residual = None
+        else:
+            assert intermediate_tensors is not None
+            hidden_states = intermediate_tensors["hidden_states"]
+            residual = intermediate_tensors["residual"]
+
+        aux_hidden_states = []
+        for idx, layer in enumerate(self.layers[self.start_layer : self.end_layer]):
+            if idx in self.aux_hidden_state_layers:
+                if residual is None:
+                    aux_hidden_states.append(hidden_states)
+                else:
+                    aux_hidden_states.append(hidden_states + residual)
+            hidden_states, residual = layer(positions, hidden_states, residual)
+
+        if not get_pp_group().is_last_rank:
+            return IntermediateTensors(
+                {"hidden_states": hidden_states, "residual": residual}
+            )
+
+        hidden_states, _ = self.norm(hidden_states, residual)
+        if aux_hidden_states:
+            return hidden_states, aux_hidden_states
+        return hidden_states
+
+
+class Step1ForCausalLM(nn.Module, SupportsPP):
+    packed_modules_mapping = STEP_PACKED_MODULES_MAPPING
+
+    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
+        super().__init__()
+        config = vllm_config.model_config.hf_config
+        quant_config = vllm_config.quant_config
+
+        self.config = config
+        self.quant_config = quant_config
+        self.model = StepDecoderModel(
+            vllm_config=vllm_config,
+            prefix=maybe_prefix(prefix, "model"),
+        )
+
+        if get_pp_group().is_last_rank:
+            self.lm_head = ParallelLMHead(
+                config.vocab_size,
+                config.hidden_size,
+                quant_config=quant_config,
+                prefix=maybe_prefix(prefix, "lm_head"),
+            )
+            if getattr(config, "tie_word_embeddings", True):
+                self.lm_head = self.lm_head.tie_weights(self.model.embed_tokens)
+            self.logits_processor = LogitsProcessor(config.vocab_size)
+        else:
+            self.lm_head = PPMissingLayer()
+            self.logits_processor = None  # type: ignore[assignment]
+
+        self.make_empty_intermediate_tensors = (
+            self.model.make_empty_intermediate_tensors
+        )
+
+    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.embed_input_ids(input_ids)
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None,
+        positions: torch.Tensor,
+        intermediate_tensors: IntermediateTensors | None,
+        inputs_embeds: torch.Tensor | None = None,
+    ) -> torch.Tensor | IntermediateTensors | tuple[torch.Tensor, list[torch.Tensor]]:
+        return self.model(
+            input_ids,
+            positions,
+            intermediate_tensors,
+            inputs_embeds=inputs_embeds,
+        )
+
+    def compute_logits(
+        self,
+        hidden_states: torch.Tensor,
+    ) -> torch.Tensor | None:
+        if not get_pp_group().is_last_rank:
+            return None
+        return self.logits_processor(self.lm_head, hidden_states)
+
+    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
+        loader = AutoWeightsLoader(self)
+        return loader.load_weights(weights)

vllm/v1/attention/backend.py

@@ -172,6 +172,10 @@ class AttentionBackend(ABC):
     def supports_sink(cls) -> bool:
         return False
 
+    @classmethod
+    def supports_alibi_sqrt(cls) -> bool:
+        return False
+
     @classmethod
     def supports_mm_prefix(cls) -> bool:
         return False

vllm/v1/attention/backends/triton_attn.py

@@ -331,6 +331,10 @@ class TritonAttentionBackend(AttentionBackend):
             AttentionType.ENCODER_DECODER,
         )
 
+    @classmethod
+    def supports_alibi_sqrt(cls) -> bool:
+        return True
+
     @classmethod
     def supports_compute_capability(cls, capability: DeviceCapability) -> bool:
         return True
@@ -353,6 +357,7 @@ class TritonAttentionImpl(AttentionImpl):
         attn_type: AttentionType = AttentionType.DECODER,
         kv_sharing_target_layer_name: int | None = None,
         sinks: torch.Tensor | None = None,
+        use_alibi_sqrt: bool = False,
     ) -> None:
         self.num_heads = num_heads
         self.head_size = head_size
@@ -386,7 +391,7 @@ class TritonAttentionImpl(AttentionImpl):
                 f"heads in the layer. Sinks shape: {sinks.shape}, "
                 f"num_heads: {num_heads}."
             )
-
+        self.use_alibi_sqrt = use_alibi_sqrt
         self.supports_quant_query_input = current_platform.is_cuda()
 
     def forward(
@@ -513,6 +518,7 @@ class TritonAttentionImpl(AttentionImpl):
             softmax_scale=self.scale,
             causal=True,
             alibi_slopes=self.alibi_slopes,
+            use_alibi_sqrt=self.use_alibi_sqrt,
             window_size=self.sliding_window,
             block_table=block_table,
             softcap=self.logits_soft_cap,

vllm/v1/attention/ops/triton_unified_attention.py

@@ -82,6 +82,7 @@ def kernel_unified_attention_2d(
     HEAD_SIZE: tl.constexpr,  # int
     HEAD_SIZE_PADDED: tl.constexpr,  # int, must be power of 2
     USE_ALIBI_SLOPES: tl.constexpr,  # bool
+    USE_ALIBI_SQRT: tl.constexpr,  # bool
     USE_QQ_BIAS: tl.constexpr,  # bool
     USE_SOFTCAP: tl.constexpr,  # bool
     USE_SINKS: tl.constexpr,  # bool
@@ -325,7 +326,16 @@ def kernel_unified_attention_2d(
         )
 
         if USE_ALIBI_SLOPES:
-            S += alibi_slope[:, None] * (seq_offset - context_len)
+            if USE_ALIBI_SQRT:
+                relative_pos = seq_offset - (context_len + query_pos[:, None])
+                alibi_offset = tl.where(
+                    relative_pos <= 0,
+                    -tl.sqrt((-relative_pos).to(tl.float32)),
+                    0.0,
+                )
+            else:
+                alibi_offset = seq_offset - context_len
+            S += alibi_slope[:, None] * alibi_offset
 
         if USE_QQ_BIAS:
             # compute key positions relative to query section
@@ -420,6 +430,7 @@ def kernel_unified_attention_3d(
     HEAD_SIZE: tl.constexpr,  # int
     HEAD_SIZE_PADDED: tl.constexpr,  # int, must be power of 2
     USE_ALIBI_SLOPES: tl.constexpr,  # bool
+    USE_ALIBI_SQRT: tl.constexpr,  # bool
     USE_QQ_BIAS: tl.constexpr,  # bool
     USE_SOFTCAP: tl.constexpr,  # bool
     USE_SINKS: tl.constexpr,  # bool
@@ -669,7 +680,16 @@ def kernel_unified_attention_3d(
         )
 
         if USE_ALIBI_SLOPES:
-            S += alibi_slope[:, None] * (seq_offset - context_len)
+            if USE_ALIBI_SQRT:
+                relative_pos = seq_offset - (context_len + query_pos[:, None])
+                alibi_offset = tl.where(
+                    relative_pos <= 0,
+                    -tl.sqrt((-relative_pos).to(tl.float32)),
+                    0.0,
+                )
+            else:
+                alibi_offset = seq_offset - context_len
+            S += alibi_slope[:, None] * alibi_offset
 
         if USE_QQ_BIAS:
             # compute key positions relative to query section
@@ -888,6 +908,7 @@ def unified_attention(
     sinks=None,
     # Optional tensor for prefix lengths (PrefixLM support)
     mm_prefix_range=None,
+    use_alibi_sqrt=False,
 ):
     assert causal, "Only causal attention is supported"
     assert q_descale is None, "Q scales not supported"
@@ -994,6 +1015,7 @@ def unified_attention(
             HEAD_SIZE=head_size,
             HEAD_SIZE_PADDED=triton.next_power_of_2(head_size),
             USE_ALIBI_SLOPES=use_alibi_slopes,
+            USE_ALIBI_SQRT=use_alibi_sqrt,
             USE_QQ_BIAS=use_qq_bias,
             USE_SOFTCAP=(softcap > 0),
             USE_SINKS=(sinks is not None),
@@ -1045,6 +1067,7 @@ def unified_attention(
             HEAD_SIZE=head_size,
             HEAD_SIZE_PADDED=triton.next_power_of_2(head_size),
             USE_ALIBI_SLOPES=use_alibi_slopes,
+            USE_ALIBI_SQRT=use_alibi_sqrt,
             USE_QQ_BIAS=use_qq_bias,
             USE_SOFTCAP=(softcap > 0),
             USE_SINKS=(sinks is not None),