Instructions to use AbstractPhil/mini-beatrix-1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AbstractPhil/mini-beatrix-1 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="AbstractPhil/mini-beatrix-1", trust_remote_code=True)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("AbstractPhil/mini-beatrix-1", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use AbstractPhil/mini-beatrix-1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AbstractPhil/mini-beatrix-1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AbstractPhil/mini-beatrix-1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/AbstractPhil/mini-beatrix-1

SGLang

How to use AbstractPhil/mini-beatrix-1 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "AbstractPhil/mini-beatrix-1" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AbstractPhil/mini-beatrix-1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "AbstractPhil/mini-beatrix-1" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AbstractPhil/mini-beatrix-1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use AbstractPhil/mini-beatrix-1 with Docker Model Runner:
```
docker model run hf.co/AbstractPhil/mini-beatrix-1
```

AbstractPhil commited on Aug 26

Commit

64f2a91

verified ·

1 Parent(s): 24c682b

mirror alephlm.py from geolip 0.7.0 @0c946b3 (hub_ckpt; default 0 = v1 verbatim)

Browse files

Files changed (1) hide show

alephlm.py +17 -1

alephlm.py CHANGED Viewed

@@ -15,6 +15,9 @@ from typing import NamedTuple, Optional
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 class LMOutput(NamedTuple):
@@ -36,6 +39,7 @@ class Block(nn.Module):
         super().__init__()
         d = cfg.d_model
         self.is_hub = layer_idx in cfg.hub_layers
         self.n1 = nn.LayerNorm(d)
         self.n2 = nn.LayerNorm(d)
         if self.is_hub:
@@ -48,8 +52,20 @@ class Block(nn.Module):
                                  cfg.gate_init)
     def forward(self, x, disable_bank=False, disable_hub=False):
         if not (disable_hub and self.is_hub):
-            x = x + self.attn(self.n1(x))
         return x + self.bank(self.n2(x), disable_dispatch=disable_bank)
     def prefill(self, x):

 import torch
 import torch.nn as nn
 import torch.nn.functional as F
+import torch.utils.checkpoint  # noqa: F401 — torch<=2.4 does NOT auto-import
+# this submodule; without it the hub_ckpt path AttributeErrors at step 1
+# (the recorded A40 landmine, 2026-08-06)
 class LMOutput(NamedTuple):
         super().__init__()
         d = cfg.d_model
         self.is_hub = layer_idx in cfg.hub_layers
+        self.ckpt_mode = getattr(cfg, "hub_ckpt", 0)
         self.n1 = nn.LayerNorm(d)
         self.n2 = nn.LayerNorm(d)
         if self.is_hub:
                                  cfg.gate_init)
     def forward(self, x, disable_bank=False, disable_hub=False):
+        # hub_ckpt (2026-08-26): recompute-in-backward for the heavy
+        # branches — at v2 scale the retained scan tensors alone exceed a
+        # 95GB card (measured OOM). Training-path only; eval/decode and
+        # the census replay (which calls attn/bank directly) are untouched.
+        ckpt = self.ckpt_mode and self.training and torch.is_grad_enabled()
         if not (disable_hub and self.is_hub):
+            if ckpt:
+                x = x + torch.utils.checkpoint.checkpoint(
+                    lambda t: self.attn(t), self.n1(x), use_reentrant=False)
+            else:
+                x = x + self.attn(self.n1(x))
+        if ckpt and self.ckpt_mode >= 2 and not disable_bank:
+            return x + torch.utils.checkpoint.checkpoint(
+                lambda t: self.bank(t), self.n2(x), use_reentrant=False)
         return x + self.bank(self.n2(x), disable_dispatch=disable_bank)
     def prefill(self, x):