Instructions to use bluemorpholimited/morphoalgo-usstock-14b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use bluemorpholimited/morphoalgo-usstock-14b with Transformers:

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

pipe = pipeline("text-generation", model="bluemorpholimited/morphoalgo-usstock-14b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("bluemorpholimited/morphoalgo-usstock-14b")
model = AutoModelForCausalLM.from_pretrained("bluemorpholimited/morphoalgo-usstock-14b", 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 bluemorpholimited/morphoalgo-usstock-14b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "bluemorpholimited/morphoalgo-usstock-14b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bluemorpholimited/morphoalgo-usstock-14b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/bluemorpholimited/morphoalgo-usstock-14b

SGLang

How to use bluemorpholimited/morphoalgo-usstock-14b 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 "bluemorpholimited/morphoalgo-usstock-14b" \
    --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": "bluemorpholimited/morphoalgo-usstock-14b",
		"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 "bluemorpholimited/morphoalgo-usstock-14b" \
        --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": "bluemorpholimited/morphoalgo-usstock-14b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use bluemorpholimited/morphoalgo-usstock-14b with Docker Model Runner:
```
docker model run hf.co/bluemorpholimited/morphoalgo-usstock-14b
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

🦋 MorphoAlgo‑14B

The US‑Equities Quantitative Analyst

BlueMorpho Limited · Quantitative Intelligence Division

14B · US equities · fundamentals + market data · commercially licensed

✨ What is MorphoAlgo‑14B?

MorphoAlgo‑14B is a domain‑specialised large language model engineered by BlueMorpho Limited for quantitative analysis of the US stock market. It reasons over fundamentals, price history, and market structure, and answers with concrete numbers, periods, and explicit reasoning — not vague commentary.

MorphoAlgo‑14B is the result of a full, in‑house alignment pipeline developed end‑to‑end by BlueMorpho: a large‑scale quantitative pretraining corpus, a supervised instruction stage, a preference‑optimisation stage, verification, and consumer‑deployable quantised exports.

📊 Results at a glance

Stage	Key metric	Result
SFT	steps · epochs · supervised tokens	6,126 · 2 · 29,551,965
SFT	final eval loss · token accuracy	0.334 · 88.2 %
DPO	steps · preference pairs	1,583 · 25,844
DPO	eval loss · reward accuracy · reward margin	4.58 e‑05 · 100 % · 13.33
Merge	artefact	14B dense (bf16)
Verification	grounding tests passed	3 / 3
GGUF	f16 · Q5_K_M · Q4_K_M	29.54 · 10.51 · 9.00 GB
Distill	student	3B (in progress)

🧠 The full training pipeline

1️⃣ Stage 2 — Supervised Fine‑Tuning

The model is fine‑tuned to behave as a disciplined quantitative analyst: it reads a market question, selects the relevant metrics, and answers with grounded figures.

Metric	Value
Optimization steps	6,126
Epochs	2
Supervised tokens	29,551,965
Final eval loss	0.334
Final token accuracy	88.2 %
Eval entropy	0.330
Best checkpoint	step 6,126

The training loss falls smoothly from ~2.8 to ~0.33 and the eval loss tracks it without divergence — a clean, well‑conditioned supervised fit.

2️⃣ Stage 3 — Direct Preference Optimisation

A calibrated DPO stage sharpens answer quality, teaching the model to prefer truthful, precisely‑scoped analysis over plausible‑but‑wrong commentary.

Metric	Start (step 200)	End (step 1,583)
Reward margin	10.61	13.33
Eval loss	5.51 e‑04	4.58 e‑05
Reward accuracy	100 %	100 %
Token accuracy	73.9 %	74.0 %
Entropy	0.460	0.434
Chosen reward	−4.10	−5.63
Rejected reward	−14.71	−18.96

Full evaluation trajectory

All eight evaluation checkpoints, straight from the training log:

The preference margin climbs monotonically while policy entropy decreases only slightly (0.460 → 0.434) — the signature of a healthy, non‑collapsing DPO run. The model learns to separate good from bad answers without degenerating into a brittle, low‑entropy policy.

📚 Data at the heart of MorphoAlgo

MorphoAlgo‑14B is trained on two purpose‑built datasets covering the US equities universe.

Dataset	Size	Tokens
SFT instruction corpus	98,000 train / 2,000 valid examples	29.55 M
DPO preference pairs (v2)	25,327 train / 517 valid pairs	6.02 M

The preference set is built around a deliberate error taxonomy — the five failure modes a quantitative analyst must avoid:

Failure mode	Pairs
Swapped / mis‑applied metric	7,751
Over‑confident claim	5,239
Subtle causal error	5,194
Plausible but wrong numbers	4,870
Wrong directional call	2,790

Underlying corpora: 10,426 US tickers of full daily OHLCV, 27,226,054 SEC EDGAR filing rows (1993 → 2026), and 8,020 XBRL company‑fact sets.

✅ Stage 5 — Verification

Every deployable artefact is verified before release.

Greedy grounding tests: 3/3 analyst prompts (AAPL, NVDA, MSFT) produced numeric, period‑grounded answers under do_sample=False.
GGUF integrity: all three exports carry a valid GGUF v3 header and re‑load and generate correctly on GPU.
Provenance: none of the published model cards name an upstream base model.

🎓 Stage 7 — Distillation

The distillation stage produces a lighter 3B student that retains the teacher's analytical behaviour for latency‑ and cost‑sensitive deployment.

🚀 Quick start

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "bluemorpholimited/morphoalgo-usstock-14b"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", torch_dtype="bfloat16")

sys = ("You are MorphoAlgo, an expert quantitative analyst specialising in the US stock market. "
       "Answer precisely with concrete numbers, periods, and clear reasoning.")
q = "Analyse AAPL between 2019-01-02 and 2020-01-02 and summarise the move."
msgs = [{"role": "system", "content": sys}, {"role": "user", "content": q}]
ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(ids, max_new_tokens=256, do_sample=False)
print(tok.decode(out[0][ids.shape[-1]:], skip_special_tokens=True))

Quantised / consumer deployment

For GGUF builds (llama.cpp, Ollama, LM Studio, llama‑cpp‑python), see bluemorpholimited/morphoalgo-usstock-14b-gguf.

📦 Model family

Repository	What it is
`morphoalgo-usstock-14b`	merged, deployable model (this repo)
`morphoalgo-usstock-14b-sft`	SFT stage adapter
`morphoalgo-usstock-14b-dpo`	DPO stage adapter
`morphoalgo-usstock-14b-gguf`	GGUF f16 / Q5_K_M / Q4_K_M
`morphoalgo-usstock-dataset`	SFT + DPO training datasets

⚠️ Limitations & responsible use

Not investment advice. MorphoAlgo‑14B produces historical, analytical commentary. It does not predict markets and must not be used as the sole basis for trading decisions.
Verify before acting. Always validate any figure against a primary market‑data source.
US‑listed equities focus. Coverage of OTC, illiquid, or newly listed instruments is limited.

📜 Licence

Released under the BlueMorpho Commercial Licence. For licensing, evaluation, and enterprise deployment support, contact the BlueMorpho Limited Quantitative Intelligence Division.

BlueMorpho Limited · Quantitative Intelligence Division

Precision. Grounding. Market intelligence.

Downloads last month: 5

Safetensors

Model size

15B params

Tensor type

BF16

Model tree for bluemorpholimited/morphoalgo-usstock-14b

Quantizations

1 model