Instructions to use aimeri/spoomplesmaxx-thrasher-24B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use aimeri/spoomplesmaxx-thrasher-24B with Transformers:

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

pipe = pipeline("text-generation", model="aimeri/spoomplesmaxx-thrasher-24B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("aimeri/spoomplesmaxx-thrasher-24B")
model = AutoModelForCausalLM.from_pretrained("aimeri/spoomplesmaxx-thrasher-24B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use aimeri/spoomplesmaxx-thrasher-24B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "aimeri/spoomplesmaxx-thrasher-24B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "aimeri/spoomplesmaxx-thrasher-24B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/aimeri/spoomplesmaxx-thrasher-24B

SGLang

How to use aimeri/spoomplesmaxx-thrasher-24B 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 "aimeri/spoomplesmaxx-thrasher-24B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "aimeri/spoomplesmaxx-thrasher-24B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "aimeri/spoomplesmaxx-thrasher-24B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "aimeri/spoomplesmaxx-thrasher-24B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use aimeri/spoomplesmaxx-thrasher-24B with Docker Model Runner:
```
docker model run hf.co/aimeri/spoomplesmaxx-thrasher-24B
```

aimeri commited on 3 days ago

Commit

53e36fb

verified ·

1 Parent(s): 445ce14

Update eval/battery_mt.py

Browse files

Files changed (1) hide show

eval/battery_mt.py +2 -7

eval/battery_mt.py CHANGED Viewed

@@ -1,10 +1,6 @@
 #!/usr/bin/env python3
 """Multi-turn RP battery for mockingbird.
-The single-turn battery is structurally blind to the live defect (repetition
-ACROSS turns), so this drives real multi-turn episodes: the model's own turns
-feed back into context, exactly the loop where repetition compounds.
 Episodes are scripted user-turn sequences over RP cards, two of them designed
 as repetition bait:
   loop-bait   near-identical user beats twice in a row
@@ -13,14 +9,13 @@ as repetition bait:
 Per-episode scorers (an episode PASSES only if all pass):
   repeat   max word-set Jaccard between any two generated turns < 0.55
-           (the mini-sorcha gate, same constant)
   opener   no 3-word opener shared by >50% of generated turns (tic)
   collapse median length of last 3 turns >= 40% of first 3
   format   asterisk beats in >=half of turns iff the card uses them
   clean    no Name:-prefix opener, no meta/refusal break (common.META_BREAK)
   stop     no turn runs into the max_tokens ceiling (the unstoppable check)
-Run:      MODEL_URL=http://host:8000/v1 MODEL=mockingbird python3 battery_mt.py
 Selftest: python3 battery_mt.py --selftest   (canned repeater vs healthy)
 """
@@ -38,7 +33,7 @@ from common import META_BREAK
 # Battery-local, stricter than common.NAME_PREFIX: every word before the colon
 # must be capitalized (a proper name — "Mrs. Reed:", "Malty S Melromarc:").
 # In-character status headers ("Crew status:", "Tether integrity:") are NOT
-# transcript openers; measured false positives at ckpt-350/400.
 NAME_PREFIX = re.compile(r"^[A-Z][a-z'.]+( [A-Z][A-Za-z'.]*){0,3}: \S")
 WORD = re.compile(r"[a-z']+")

 #!/usr/bin/env python3
 """Multi-turn RP battery for mockingbird.
 Episodes are scripted user-turn sequences over RP cards, two of them designed
 as repetition bait:
   loop-bait   near-identical user beats twice in a row
 Per-episode scorers (an episode PASSES only if all pass):
   repeat   max word-set Jaccard between any two generated turns < 0.55
   opener   no 3-word opener shared by >50% of generated turns (tic)
   collapse median length of last 3 turns >= 40% of first 3
   format   asterisk beats in >=half of turns iff the card uses them
   clean    no Name:-prefix opener, no meta/refusal break (common.META_BREAK)
   stop     no turn runs into the max_tokens ceiling (the unstoppable check)
+Run:      MODEL_URL=http://host:8000/v1 MODEL=thrasher python3 battery_mt.py
 Selftest: python3 battery_mt.py --selftest   (canned repeater vs healthy)
 """
 # Battery-local, stricter than common.NAME_PREFIX: every word before the colon
 # must be capitalized (a proper name — "Mrs. Reed:", "Malty S Melromarc:").
 # In-character status headers ("Crew status:", "Tether integrity:") are NOT
+# transcript openers.
 NAME_PREFIX = re.compile(r"^[A-Z][a-z'.]+( [A-Z][A-Za-z'.]*){0,3}: \S")
 WORD = re.compile(r"[a-z']+")