Upload folder using huggingface_hub

.gitattributes

@@ -72,3 +72,4 @@ checkpoints/qwen3_hyp/qwen3_hyp_p1083m_t1_4_suite-owt_scaling_v3_family-qwen3_ge
 checkpoints/qwen3_hyp/qwen3_hyp_p2324m_t1_128_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt3_20260209_141700/checkpoint-251/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 checkpoints/qwen3_hyp/qwen3_hyp_p407m_t1_1_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260312_143830/checkpoint-8192/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_2_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260219_093238/checkpoint-8192/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/tokenizer.json filter=lfs diff=lfs merge=lfs -text

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/added_tokens.json

@@ -0,0 +1,28 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|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
+}

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/config.json

@@ -0,0 +1,58 @@
+{
+  "architectures": [
+    "Qwen3HypForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoModel": "modeling_qwen3_hyp.Qwen3HypForCausalLM",
+    "AutoModelForCausalLM": "modeling_qwen3_hyp.Qwen3HypForCausalLM"
+  },
+  "bos_token_id": 151643,
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 1280,
+  "initializer_range": 0.02,
+  "intermediate_size": 3840,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "lorentz_k0": 1.0,
+  "max_position_embeddings": 40960,
+  "max_window_layers": 20,
+  "model_type": "qwen3",
+  "num_attention_heads": 20,
+  "num_hidden_layers": 20,
+  "num_key_value_heads": 10,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "transformers_version": "4.51.3",
+  "use_cache": false
+}

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/global_step3072/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e89eb6b32bc55283c67111da9d49e92945068c2417206237d7b8388c963954ed
+size 3225502960

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/global_step3072/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b94cdefb9e1d395b0ab7682c51a98341e57031fbcff4bf86403f77f1d178c879
+size 3225508336

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/global_step3072/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57bc87c0c45250d8c03704eae628078c18bac54a932f7abd3a8c8ea6e6b9ed97
+size 3225505072

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/global_step3072/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a75e92ba6d057dea61b59bfce03e0701a0db19ea4952e017b81ec77d449f62e8
+size 3225500912

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/global_step3072/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8d2f762ef273ca8996bdb963b66569ba67d005e56e13e0f3ad8706c255c1f1a
+size 2150395192

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/latest

@@ -0,0 +1 @@
+global_step3072

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e483654b326b07eb11bbf76301ff0bff5da570ba84b4adc71d4c1bd6db9832ce
+size 2539616244

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/modeling_qwen3_hyp.py

@@ -0,0 +1,182 @@
+"""
+Hyperbolic-tail adapter for Qwen3.
+
+This wrapper keeps the original Qwen3 stack unchanged and only projects the
+final hidden state onto a Lorentz hyperboloid before the vocabulary projection.
+The projection uses a learnable curvature (softplus-parameterized) and a
+learnable per-dimension scaling to stabilize norms. A new lm_head is created
+with an extra time dimension; existing weights are copied into the spatial
+coordinates and the time coordinate is initialized to zero.
+"""
+
+import math
+from typing import Any, Optional
+
+import torch
+import torch.distributed as dist
+import torch.nn.functional as F
+from transformers.modeling_outputs import CausalLMOutputWithPast
+from transformers.models.qwen3.modeling_qwen3 import Qwen3ForCausalLM
+
+
+class Qwen3HypForCausalLM(Qwen3ForCausalLM):
+    """
+    Qwen3 with a hyperbolic (Lorentz) projection applied only at the output.
+    Fixed: Uses correct Exponential Map (cosh) and Minkowski Inner Product.
+    Includes Logit Scaling to match initialization variance with Euclidean baseline.
+    """
+
+    def __init__(self, config: Any):
+        super().__init__(config)
+
+        hidden = self.lm_head.weight.shape[1]
+        vocab = self.lm_head.weight.shape[0]
+
+        # 1. 曲率参数 (Learnable curvature)
+        k0 = getattr(config, "lorentz_k0", 1.0)
+        self.k_param = torch.nn.Parameter(torch.log(torch.expm1(torch.tensor(float(k0)))))
+
+        # 2. 模长缩放 (Norm Scaling)
+        # 初始化为 sqrt(hidden) 是个很好的 trick，能让初始输入模长接近 1.0
+        init_scale = math.sqrt(hidden)
+        self.lora_norm_scale = torch.nn.Parameter(torch.full((hidden,), init_scale))
+
+        # 3. Logit Scale (Compensation for Hyperbolic Projection norm reduction)
+        # 目标：让双曲输入的能量与欧氏 RMSNorm 后的能量一致
+        # Ratio = sqrt(D) / sinh(1.0)
+        init_scale_factor = math.sqrt(hidden) / math.sinh(1.0)
+        # 使用 log 参数化是为了数值稳定，保证 scale 恒为正
+        self.logit_scale = torch.nn.Parameter(torch.tensor(math.log(init_scale_factor)))
+
+        # 4. 备份与初始化 Head
+        old_w = self.lm_head.weight.data
+        self.euclidean_head = torch.nn.Linear(hidden, vocab, bias=False)
+        self.euclidean_head.weight.data.copy_(old_w)
+
+        # Hyperbolic Head
+        # 我们使用 Linear 实现闵可夫斯基内积，但在 forward 中会使用 Input-Flipping Trick
+        new_w = torch.zeros((vocab, hidden + 1), dtype=old_w.dtype, device=old_w.device)
+        new_w[:, 1:] = old_w
+        # 时间维权重初始化为0：初始状态下 <x, w>_L = -0*x0 + ws*xs = Euclidean Dot Product
+        # 这保证了模型从欧氏空间平滑过渡。
+        self.hyper_head = torch.nn.Linear(hidden + 1, vocab, bias=False)
+        self.hyper_head.weight.data.copy_(new_w)
+
+        self.lm_head = self.hyper_head
+
+    def _project_hyperbolic(self, x: torch.Tensor) -> torch.Tensor:
+        """
+        Correct Exponential Map: Tangent Space -> Lorentz Hyperboloid
+        """
+        # [Stability] 在 Scaling 实验中，双曲函数极易溢出，建议强制 float32
+        orig_dtype = x.dtype
+        x = x.float()
+
+        # 1. 计算曲率与归一化
+        k_pos = torch.nn.functional.softplus(self.k_param) + 1e-6
+        sqrt_k = torch.sqrt(k_pos)
+
+        # 归一化输入 (Input is v in tangent space)
+        x_scaled = x / self.lora_norm_scale.float()
+        x_norm_scaled = torch.norm(x_scaled, p=2, dim=-1, keepdim=True)
+
+        # 2. 计算归一化半径 r / sqrt(k)
+        theta = x_norm_scaled / sqrt_k
+        
+        # [Stability] Clamp 防止 exp 溢出 (exp(10) ~ 22000, BF16 安全范围)
+        theta = torch.clamp(theta, max=10.0)
+
+        # 3. 指数映射 (Exponential Map)
+        # Time: sqrt(k) * cosh(theta)  <-- 修正了 arccosh
+        x_time = sqrt_k * torch.cosh(theta)
+
+        # Space: sqrt(k) * sinh(theta) * (v / |v|)
+        # 加上 1e-7 防止除零
+        x_dir = x_scaled / x_norm_scaled.clamp(min=1e-7)
+        x_space = sqrt_k * torch.sinh(theta) * x_dir
+
+        # 拼接并还原精度
+        out = torch.cat([x_time, x_space], dim=-1)
+        return out.to(orig_dtype)
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        **kwargs: Any,
+    ):
+        # 显式传入 return_dict=True，返回的是对象
+        transformer_outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            output_hidden_states=True,
+            return_dict=True,
+            **kwargs,
+        )
+
+        h = transformer_outputs.hidden_states[-1]
+        
+        # 1. 投影到双曲面
+        h_hyp = self._project_hyperbolic(h)
+
+        # [Monitor] 记录 curvature k
+        if self.training:
+            k_val = torch.nn.functional.softplus(self.k_param) + 1e-6
+            should_log = False
+            if dist.is_initialized():
+                if dist.get_rank() == 0:
+                    should_log = True
+            else:
+                should_log = True
+            
+            if should_log:
+                try:
+                    import wandb
+                    if wandb.run is not None:
+                        wandb.log({"train/hyp_k": k_val.item()}, commit=False)
+                except ImportError:
+                    pass
+
+        # 2. 计算 Logits (Minkowski Inner Product Trick)
+        h_hyp_minkowski = h_hyp.clone()
+        h_hyp_minkowski[..., 0] = -h_hyp_minkowski[..., 0]
+        
+        # [Logit Scale] 关键修正：把“消失的50倍”乘回来
+        logits = self.hyper_head(h_hyp_minkowski) * self.logit_scale.exp()
+
+        loss = None
+        if labels is not None:
+            # Shift for next-token prediction
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            
+            # [Fix Loss Display] 使用 sum reduction 并手动除以 num_items_in_batch
+            loss = F.cross_entropy(
+                shift_logits.view(-1, shift_logits.size(-1)),
+                shift_labels.view(-1),
+                ignore_index=-100,
+                reduction='sum'
+            )
+            
+            num_items = kwargs.get("num_items_in_batch")
+            if num_items is not None:
+                loss = loss / num_items
+            else:
+                valid_items = (shift_labels != -100).sum()
+                if valid_items > 0:
+                    loss = loss / valid_items
+                else:
+                    loss = loss / shift_labels.numel()
+
+        if not kwargs.get("return_dict", True):
+            output = (logits,) + (transformer_outputs.past_key_values, transformer_outputs.hidden_states, transformer_outputs.attentions)
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef002048764051a71fb00f8f978e9ec32b780dc850bdb059af362cc56494234b
+size 15024

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37194a6d48612e1a46a2d5d317ead97c70d9fc4569b0118fcd5f84c3dc9daa5a
+size 15024

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17c179483659a784aa1ace2427daff48c556a6bcc3c330e6f3274e4dc95e4b49
+size 15024

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b56857c9b117629f35af2c3d64f522d33a9d8aa94faa81ec6956380a895118c4
+size 15024

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2819acc651556e7afe4455bcce325e923c7d86faf0bd0546f95f11de8f251294
+size 1064

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/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": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
+size 11422654

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/tokenizer_config.json

@@ -0,0 +1,241 @@
+{
+  "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
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "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,
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0].role == 'system' %}\n        {{- messages[0].content + '\\n\\n' }}\n    {%- endif %}\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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0].role == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0].content + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}\n        {%- set ns.multi_step_tool = false %}\n        {%- set ns.last_query_index = index %}\n    {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n    {%- if message.content is string %}\n        {%- set content = message.content %}\n    {%- else %}\n        {%- set content = '' %}\n    {%- endif %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) %}\n        {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is string %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- else %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n                {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- if loop.index0 > ns.last_query_index %}\n            {%- if loop.last or (not loop.last and reasoning_content) %}\n                {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content.strip('\\n') + '\\n</think>\\n\\n' + content.lstrip('\\n') }}\n            {%- else %}\n                {{- '<|im_start|>' + message.role + '\\n' + content }}\n            {%- endif %}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if (loop.first and content) or (not loop.first) %}\n                    {{- '\\n' }}\n                {%- endif %}\n                {%- if tool_call.function %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {{- '<tool_call>\\n{\"name\": \"' }}\n                {{- tool_call.name }}\n                {{- '\", \"arguments\": ' }}\n                {%- if tool_call.arguments is string %}\n                    {{- tool_call.arguments }}\n                {%- else %}\n                    {{- tool_call.arguments | tojson }}\n                {%- endif %}\n                {{- '}\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n    {%- if enable_thinking is defined and enable_thinking is false %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/trainer_state.json

@@ -0,0 +1,292 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.0924177299443824,
+  "eval_steps": 512,
+  "global_step": 3072,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.003008389646627031,
+      "grad_norm": 1.4079504013061523,
+      "learning_rate": 6.149068322981368e-05,
+      "loss": 9.3539,
+      "step": 100
+    },
+    {
+      "epoch": 0.006016779293254062,
+      "grad_norm": 1.252025842666626,
+      "learning_rate": 9.999421087841762e-05,
+      "loss": 6.6429,
+      "step": 200
+    },
+    {
+      "epoch": 0.009025168939881093,
+      "grad_norm": 0.8464359045028687,
+      "learning_rate": 9.992366890377964e-05,
+      "loss": 5.8777,
+      "step": 300
+    },
+    {
+      "epoch": 0.012033558586508125,
+      "grad_norm": 0.8979347944259644,
+      "learning_rate": 9.9773076750685e-05,
+      "loss": 5.4891,
+      "step": 400
+    },
+    {
+      "epoch": 0.015041948233135156,
+      "grad_norm": 0.7227486968040466,
+      "learning_rate": 9.954267588633746e-05,
+      "loss": 5.2154,
+      "step": 500
+    },
+    {
+      "epoch": 0.015402954990730399,
+      "eval_loss": 5.0861992835998535,
+      "eval_runtime": 28.6174,
+      "eval_samples_per_second": 148.092,
+      "eval_steps_per_second": 4.648,
+      "step": 512
+    },
+    {
+      "epoch": 0.018050337879762186,
+      "grad_norm": 0.6411813497543335,
+      "learning_rate": 9.923283574733313e-05,
+      "loss": 5.0171,
+      "step": 600
+    },
+    {
+      "epoch": 0.021058727526389216,
+      "grad_norm": 0.6219043731689453,
+      "learning_rate": 9.8844053147287e-05,
+      "loss": 4.8501,
+      "step": 700
+    },
+    {
+      "epoch": 0.02406711717301625,
+      "grad_norm": 0.6222556829452515,
+      "learning_rate": 9.837695148021635e-05,
+      "loss": 4.7037,
+      "step": 800
+    },
+    {
+      "epoch": 0.02707550681964328,
+      "grad_norm": 0.6238115429878235,
+      "learning_rate": 9.78322797209587e-05,
+      "loss": 4.5693,
+      "step": 900
+    },
+    {
+      "epoch": 0.030083896466270313,
+      "grad_norm": 0.6507419943809509,
+      "learning_rate": 9.721091122422707e-05,
+      "loss": 4.4618,
+      "step": 1000
+    },
+    {
+      "epoch": 0.030805909981460797,
+      "eval_loss": 4.389865398406982,
+      "eval_runtime": 28.5812,
+      "eval_samples_per_second": 148.279,
+      "eval_steps_per_second": 4.653,
+      "step": 1024
+    },
+    {
+      "epoch": 0.03309228611289734,
+      "grad_norm": 0.5398547053337097,
+      "learning_rate": 9.651384232422814e-05,
+      "loss": 4.3667,
+      "step": 1100
+    },
+    {
+      "epoch": 0.03610067575952437,
+      "grad_norm": 0.6056393980979919,
+      "learning_rate": 9.574219073708867e-05,
+      "loss": 4.2899,
+      "step": 1200
+    },
+    {
+      "epoch": 0.0391090654061514,
+      "grad_norm": 0.5178854465484619,
+      "learning_rate": 9.489719376865203e-05,
+      "loss": 4.221,
+      "step": 1300
+    },
+    {
+      "epoch": 0.04211745505277843,
+      "grad_norm": 0.4748992323875427,
+      "learning_rate": 9.398020633051834e-05,
+      "loss": 4.1702,
+      "step": 1400
+    },
+    {
+      "epoch": 0.04512584469940547,
+      "grad_norm": 0.5286852121353149,
+      "learning_rate": 9.299269876750955e-05,
+      "loss": 4.1169,
+      "step": 1500
+    },
+    {
+      "epoch": 0.0462088649721912,
+      "eval_loss": 4.072985649108887,
+      "eval_runtime": 28.6335,
+      "eval_samples_per_second": 148.009,
+      "eval_steps_per_second": 4.645,
+      "step": 1536
+    },
+    {
+      "epoch": 0.0481342343460325,
+      "grad_norm": 0.48330968618392944,
+      "learning_rate": 9.19362545000429e-05,
+      "loss": 4.0688,
+      "step": 1600
+    },
+    {
+      "epoch": 0.05114262399265953,
+      "grad_norm": 0.4792095422744751,
+      "learning_rate": 9.08125674851932e-05,
+      "loss": 4.0287,
+      "step": 1700
+    },
+    {
+      "epoch": 0.05415101363928656,
+      "grad_norm": 0.5017923712730408,
+      "learning_rate": 8.962343950051497e-05,
+      "loss": 3.9919,
+      "step": 1800
+    },
+    {
+      "epoch": 0.05715940328591359,
+      "grad_norm": 0.45991232991218567,
+      "learning_rate": 8.83707772549795e-05,
+      "loss": 3.959,
+      "step": 1900
+    },
+    {
+      "epoch": 0.060167792932540626,
+      "grad_norm": 0.45571720600128174,
+      "learning_rate": 8.705658933165981e-05,
+      "loss": 3.9288,
+      "step": 2000
+    },
+    {
+      "epoch": 0.061611819962921595,
+      "eval_loss": 3.8978145122528076,
+      "eval_runtime": 28.7591,
+      "eval_samples_per_second": 147.362,
+      "eval_steps_per_second": 4.625,
+      "step": 2048
+    },
+    {
+      "epoch": 0.06317618257916766,
+      "grad_norm": 0.4395545721054077,
+      "learning_rate": 8.568298296706507e-05,
+      "loss": 3.8941,
+      "step": 2100
+    },
+    {
+      "epoch": 0.06618457222579469,
+      "grad_norm": 0.4189292788505554,
+      "learning_rate": 8.425216067228932e-05,
+      "loss": 3.8736,
+      "step": 2200
+    },
+    {
+      "epoch": 0.06919296187242172,
+      "grad_norm": 0.4565911591053009,
+      "learning_rate": 8.276641670139183e-05,
+      "loss": 3.8446,
+      "step": 2300
+    },
+    {
+      "epoch": 0.07220135151904875,
+      "grad_norm": 0.3999106287956238,
+      "learning_rate": 8.122813337267216e-05,
+      "loss": 3.8308,
+      "step": 2400
+    },
+    {
+      "epoch": 0.07520974116567578,
+      "grad_norm": 0.4241749942302704,
+      "learning_rate": 7.963977724873853e-05,
+      "loss": 3.8009,
+      "step": 2500
+    },
+    {
+      "epoch": 0.077014774953652,
+      "eval_loss": 3.7836685180664062,
+      "eval_runtime": 28.6535,
+      "eval_samples_per_second": 147.905,
+      "eval_steps_per_second": 4.642,
+      "step": 2560
+    },
+    {
+      "epoch": 0.0782181308123028,
+      "grad_norm": 0.4040316343307495,
+      "learning_rate": 7.800389518149431e-05,
+      "loss": 3.7858,
+      "step": 2600
+    },
+    {
+      "epoch": 0.08122652045892984,
+      "grad_norm": 0.419279545545578,
+      "learning_rate": 7.632311022838487e-05,
+      "loss": 3.7617,
+      "step": 2700
+    },
+    {
+      "epoch": 0.08423491010555686,
+      "grad_norm": 0.41915661096572876,
+      "learning_rate": 7.460011744645226e-05,
+      "loss": 3.7543,
+      "step": 2800
+    },
+    {
+      "epoch": 0.08724329975218391,
+      "grad_norm": 0.40949028730392456,
+      "learning_rate": 7.283767957094212e-05,
+      "loss": 3.739,
+      "step": 2900
+    },
+    {
+      "epoch": 0.09025168939881094,
+      "grad_norm": 0.3878908157348633,
+      "learning_rate": 7.103862258539194e-05,
+      "loss": 3.7121,
+      "step": 3000
+    },
+    {
+      "epoch": 0.0924177299443824,
+      "eval_loss": 3.6991801261901855,
+      "eval_runtime": 28.6363,
+      "eval_samples_per_second": 147.994,
+      "eval_steps_per_second": 4.644,
+      "step": 3072
+    }
+  ],
+  "logging_steps": 100,
+  "max_steps": 8006,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 1024,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.2553369455650406e+18,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4561c62b4529babe6f1b9d1634ae524a3b841005c70c0e4f7ad61219e311d366
+size 7544

checkpoints/qwen3_hyp/qwen3_hyp_p686m_t1_4_suite-owt_scaling_v3_family-qwen3_geometry_id-hyp_variant-base_init_slope-std/attempt2_20260214_093947/checkpoint-3072/zero_to_fp32.py

@@ -0,0 +1,604 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+                                               args.output_file,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)