roberthein/modelname-new - AI Model Hubs Tool

Overview

modelname-new is a public model hosted by user roberthein on Replicate. The Replicate model page serves primarily as an API reference and playground: it exposes a web-based playground for iterative testing, an API reference (OpenAPI/schema) for programmatic use, example runs, and a versions history. Replicate’s page notes the model is open-source and can be run locally with Docker, but the model listing itself does not include a README or extended documentation, so consumers should inspect the model’s OpenAPI/schema before integrating. (Replicate model page: https://replicate.com/roberthein/modelname-new). On Replicate the listing reports about 233 runs, runs on Nvidia H100 hardware, and typical predictions complete in ~19 seconds. Replicate also shows an approximate cost of $0.029 per run (about 34 runs per USD), with the usual caveat that actual cost depends on inputs and runtime. The model has an explicit versions page and a specific version (ID shown on the versions view) that can be selected when creating predictions; you can programmatically fetch the model/version OpenAPI schema via Replicate’s HTTP API to learn exact input and output fields before calling the prediction endpoint (see Replicate OpenAPI docs for examples).

Key Features

• Web playground for iterative, interactive testing of model prompts and parameters
• Public API reference and OpenAPI/schema accessible via Replicate for programmatic integration
• Version history allowing selection of a specific model version for reproducible runs
• Replicate-hosted execution on Nvidia H100 with reported ~19s prediction latency
• Approximate per-run cost reported by Replicate (~$0.029); model listed as runnable locally via Docker
• Examples shown on the model page; minimal README on the listing, so inspect the schema before use

Example Usage

Example (python):

import os
import requests

# Set your Replicate API token in the environment: REPLICATE_API_TOKEN
TOKEN = os.environ.get("REPLICATE_API_TOKEN")
if not TOKEN:
    raise RuntimeError("Set REPLICATE_API_TOKEN environment variable")

headers = {
    "Authorization": f"Token {TOKEN}",
    "Content-Type": "application/json",
}

# 1) Fetch model metadata (including latest version and OpenAPI schema)
model_url = "https://api.replicate.com/v1/models/roberthein/modelname-new"
model_info = requests.get(model_url, headers=headers).json()
latest_version = model_info.get("latest_version")
print("Latest version metadata keys:", list(latest_version.keys() if latest_version else []))

# If the model exposes an OpenAPI schema, inspect the Input shape before sending a prediction
openapi_schema = latest_version.get("openapi_schema") if latest_version else None
if openapi_schema:
    input_props = openapi_schema.get("components", {}).get("schemas", {}).get("Input", {}).get("properties")
    print("Model input properties:", input_props)
else:
    print("No openapi_schema found on the model listing. Check the Replicate model page.")

# 2) Create a prediction (replace the input dict with fields from the model's schema)
# Note: 'version' should be the version id string (e.g. latest_version["id"]).
if latest_version and latest_version.get("id"):
    prediction_payload = {
        "version": latest_version["id"],
        "input": {
            # Example placeholder. Replace with actual fields from input_props, e.g. "prompt": "Hello"
            "prompt": "Hello, world"
        }
    }
    pred = requests.post("https://api.replicate.com/v1/predictions", headers=headers, json=prediction_payload).json()
    print("Prediction created:", pred)
    # Poll prediction URL in pred['urls'] or pred['id'] to get final outputs per Replicate API docs.
else:
    print("No usable version id found; confirm the model name and permissions.")

# For quick runs you can also use the replicate Python client: replicate.run("owner/model:version", input={...})
# but always check the model's input schema first to avoid malformed requests.

Pricing

Replicate reports an approximate cost of $0.029 per run for this model (about 34 runs per USD). Actual cost on Replicate varies with input size, execution time, and hardware; the listing also indicates the model can be run locally via Docker, allowing self-hosting to avoid Replicate runtime charges. Source: https://replicate.com/roberthein/modelname-new

Benchmarks

Approximate cost per run: $0.029 per run (Replicate estimate) (Source: https://replicate.com/roberthein/modelname-new)

Typical prediction latency: ≈19 seconds (on Replicate-hosted H100) (Source: https://replicate.com/roberthein/modelname-new)

Public run count: 233 runs (public page) (Source: https://replicate.com/roberthein/modelname-new)

Latest version timestamp (page): Version listed ~1 year, 4 months ago; model updated ~9 months ago (Source: https://replicate.com/roberthein/modelname-new/versions)

Key Information

• Category: Model Hubs
• Type: AI Model Hubs Tool