codemindai-api 1.0.0__cp313-none-any.whl

codemindai_api-1.0.0.dist-info/METADATA

@@ -0,0 +1,182 @@
+Metadata-Version: 2.1
+Name: codemindai-api
+Version: 1.0.0
+Summary: Run trained CodeMind code-intelligence models - from Python or as an OpenAI-compatible API server
+Requires-Python: ==3.13.*
+License: Proprietary
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Operating System :: Microsoft :: Windows
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Environment :: GPU :: NVIDIA CUDA
+Classifier: License :: Other/Proprietary License
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Dist: torch>=2.4
+Requires-Dist: safetensors>=0.4.5
+Requires-Dist: psutil>=5.9
+Requires-Dist: tree-sitter-language-pack>=1.0,<2
+Requires-Dist: pypdf>=4.0
+Provides-Extra: server
+Requires-Dist: fastapi>=0.115,<1 ; extra == "server"
+Requires-Dist: uvicorn[standard]>=0.30 ; extra == "server"
+Requires-Dist: pydantic>=2.7,<3 ; extra == "server"
+Requires-Dist: redis>=5.0 ; extra == "server"
+Requires-Dist: psycopg[binary]>=3.2 ; extra == "server"
+Description-Content-Type: text/markdown
+
+# codemindai-api
+
+Run trained **CodeMind** code-intelligence models on your own machine - from Python, or as an OpenAI-compatible API
+server for websites and apps. Windows and Linux, one or many NVIDIA GPUs.
+
+A CodeMind model folder (`Model_CodeMind/`) contains only data: weights (`safetensors`), configuration, tokenizers
+and documents. `codemindai-api` is what runs it.
+
+## Requirements
+
+| | |
+|---|---|
+| OS | Windows 10/11 (64-bit) or Linux (x86-64) |
+| Python | 3.11, 3.12 or 3.13 |
+| PyTorch | 2.4+ with CUDA (install it first - see below) |
+| GPU | NVIDIA, any number; memory per precision is in each model's README |
+
+## Install
+
+```bash
+python -m venv .venv
+# Linux:   source .venv/bin/activate
+# Windows: .venv\Scripts\activate
+
+# 1. PyTorch with CUDA (choose your system at https://pytorch.org/get-started/locally/), e.g.
+pip install torch --index-url https://download.pytorch.org/whl/cu128
+
+# 2. codemindai-api - add [server] for the HTTP API server
+pip install "codemindai-api[server]"
+
+codemindai-api info --model ./Model_CodeMind      # model, runtime, PyTorch, GPUs
+```
+
+## Use it from Python
+
+```python
+import codemindai_api
+
+model = codemindai_api.load("./Model_CodeMind")       # first GPU (CPU if none); FP8 where supported
+
+print(model.chat("What is the difference between a process and a thread?"))
+
+ans = model.code("A thread-safe LRU cache", language="python")
+print(ans.text, ans.confidence)
+
+print(model.validate("SELECT * FROM users WHERE id = ' + user_id", language="sql")["issues"])
+
+with codemindai_api.load("./Model_CodeMind", precision="bf16", device="cuda:1") as m2:   # one chosen GPU
+    for piece in m2.stream("Explain this regex: ^\\d{3}-\\d{4}$"):
+        print(piece, end="")
+
+big = codemindai_api.load("./Model_CodeMind", precision="fp8", devices="all")   # one copy on EVERY GPU
+answers = big.chat_many([f"Explain error code E{i}" for i in range(100)])  # spread over all of them
+```
+
+`load(path, precision="auto"|"fp32"|"bf16"|"fp16"|"fp8"|"fp4", device="auto"|"cuda:N"|"cpu", devices="all"|[...],
+split=False, fp8="auto", safety=True, before=None, after=None, verify=False, threads=0, warmup=True)`
+
+`Model.chat(prompt | messages=[...], system="", mode="auto"|"text"|"code", language=None, temperature=0.8, top_p=0.92,
+max_tokens=0)` → `Answer(text, mode, uncertain, confidence, reason, questions, language, language_supported, score, raw)`
+
+Also `code()`, `chat_many()`, `stream()`, `validate()`, `ground()` (exact math/physics with units), `count_tokens()`,
+`add_hook()`, `info`, `runtime`, `close()`.
+
+### Precision
+
+| `precision` | weights | use it for |
+|---|---|---|
+| `fp32` | 4 bytes | exact reference |
+| `bf16` | 2 bytes | same quality, half the memory (Ampere and newer) |
+| `fp16` | 2 bytes | GPUs without BF16 (Turing, Volta) |
+| `fp8` | 1 byte | FP8 tensor cores on RTX 40xx / Ada, Hopper, Blackwell (elsewhere FP8 storage, BF16 compute) |
+| `fp4` | ~5 bits | the smallest: 4-bit E2M1 values, one scale per 16 |
+| `auto` | | FP8 when the GPU passes a numeric self-test, else BF16; FP32 on CPU |
+
+Every choice is checked on the real GPU and falls back safely (FP4 → BF16, FP8 → BF16) instead of failing.
+
+### Your own code around every answer
+
+```python
+def before(ctx):                       # system, prompt, mode, language, temperature, top_p, max_tokens
+    if "password" in ctx["prompt"].lower():
+        raise ValueError("not on this service")       # refuses the request
+    return ctx
+
+model = codemindai_api.load("./Model_CodeMind", before=before, after=lambda ctx, a: a.text.strip())
+```
+
+The server takes the same idea from a file: `codemindai-api serve --hooks my_hooks.py` with `before_request(ctx)` and
+`after_response(ctx, result)`.
+
+## Run the API server
+
+```bash
+mkdir my-server && cd my-server
+codemindai-api init                      # .env with an admin token + Model_CodeMind/
+#   put the model files into Model_CodeMind/
+codemindai-api serve --supervise --host 0.0.0.0   # every GPU, auto-restart (default host 127.0.0.1 = local only)
+codemindai-api create-key --owner my-app
+```
+
+OpenAI-compatible: `POST /v1/chat/completions` (streaming), `/v1/completions`, `/v1/code/validate`, `/v1/ground`,
+`/v1/models`, `/v1/me`, `/v1/usage`, `/health`, `/ready`, plus `/admin/*` (keys, limits, usage, audit, workers,
+zero-downtime model reload) and `/metrics`. Any OpenAI SDK works:
+
+```python
+from openai import OpenAI
+client = OpenAI(base_url="http://localhost:8000/v1", api_key="YOUR_KEY")
+print(client.chat.completions.create(model="codemind-3.0",
+      messages=[{"role": "user", "content": "Hello"}]).choices[0].message.content)
+```
+
+The full reference ships with every model: `Model_CodeMind/API_REFERENCE.md`.
+
+## Hardware use
+
+- **Any number of GPUs:** the server starts model workers on every GPU (`--gpus auto|0,1,2,...`), sizes how many copies
+  fit per GPU from real memory use (`--workers-per-gpu`), scales up under load and down when idle, and spreads requests
+  over all of them. In Python: `devices="all"` + `chat_many()`.
+- **Precision:** `--precision` / `precision=` FP32, BF16, FP16, FP8, FP4 - self-tested on each GPU, with fallback.
+- **One big 3.x model over GPU pairs:** `codemindai_api.load(path, devices="all", split=True)`.
+- **Limits are yours:** queue, connections, output length, requests/minute, parallel requests per key - each can be
+  set, or `0` for unlimited (`codemindai-api serve --help`).
+- **CPU:** works without a GPU (slow - for testing).
+
+## Security
+
+- Model folders are data only (`safetensors`, no pickle): loading a model never runs code from it.
+- Every load checks the model files; `codemindai-api verify` checks each file's SHA-256 against `manifest.json`.
+- No internet needed to answer. The code parser downloads each language grammar once and caches it - run
+  `codemindai-api prefetch` once before taking a machine offline (`serve` does it at start).
+- Server: API keys stored only as HMAC-SHA256 hashes; admin token for `/admin`; per-key rate, token, concurrency,
+  IP and website limits; request size limits; content screening of requests and answers; secret redaction; audit log.
+- The server listens on this machine only (127.0.0.1) unless you pass `--host 0.0.0.0`.
+- Keep `.env` private; use HTTPS in front of a public server.
+
+## Commands
+
+| command | does |
+|---|---|
+| `codemindai-api init` | prepare the current folder for serving |
+| `codemindai-api serve` | run the API server |
+| `codemindai-api chat "..."` | one answer from the command line |
+| `codemindai-api info` / `verify` / `doctor` / `prefetch` | inspect, check files, full system check, download grammars |
+| `codemindai-api create-key` / `list-keys` / `revoke-key` / `gen-token` | API keys and tokens |
+
+## Troubleshooting
+
+| message | fix |
+|---|---|
+| `PyTorch sees no CUDA GPU` | install the CUDA build of PyTorch (step 1), update the NVIDIA driver |
+| `this process already runs the v2 runtime` | 2.x and 3.x models need separate processes |
+| `model folder is incomplete` | copy the whole model folder again; `codemindai-api verify` names the bad file |
+| out of GPU memory | `fp8="auto"` (default), a smaller `--workers-per-gpu`, or `devices=[...]` for 3.x |
+
+License: proprietary - see the terms that came with the model.

codemindai_api-1.0.0.dist-info/RECORD

@@ -0,0 +1,20 @@
+codemindai_api/__init__.pyc,sha256=vtC4Ay0_BnapjL8sH1pgeG4oHOTZHCs0HXi-JLXT7Vk,672
+codemindai_api/__main__.pyc,sha256=7x6yFPWmc_Z1lQ5cNPAQszcBeo0gUUHK_B2bnanhKOU,224
+codemindai_api/_paths.pyc,sha256=8f6rvTLvgo7SbThxAZKpaCRdTaHC7bm62X51ZzJpIEo,3536
+codemindai_api/_runtime/v2/CodeMind.pyc,sha256=Oq4zyQAFXNVYzZMExusHWWM2Pz1xd7-xmtzI4_8cHl4,1448665
+codemindai_api/_runtime/v2/codemind_brain_dual.pyc,sha256=ZMM5AXshxjk9ev7pbey9zvGkkQkjI0rNA19ThnRPZyQ,88924
+codemindai_api/_runtime/v2/linguistic_safety.pyc,sha256=DgOW2HZnI39qj8qrggXzOCcOH9Q3uerqA5_2dkEzmzs,68420
+codemindai_api/_runtime/v3/CodeMind.pyc,sha256=lIo50W--WuwqNpiAVmJL1_f6jo9MF4gtyVq0DxinLPs,1462091
+codemindai_api/_runtime/v3/codemind_brain_dual.pyc,sha256=TmRsYzOPQvWQgmt6K_kyX9MwvT0fVM_hp6vyWNv_1cc,94337
+codemindai_api/_runtime/v3/linguistic_safety.pyc,sha256=p1UfkINEaIZ-kvJ4WDFs1YhnXAPjfr4VGjWqpiwSZds,69504
+codemindai_api/api.pyc,sha256=WM-Nky1VS2Z10pAWy6osYA-KnX5gicmfi6spW6nIRUc,32263
+codemindai_api/cli.pyc,sha256=1CsDj804D_AQm7p5UyhDkLgntaMQPTLZ8CnCJmQXfXM,19816
+codemindai_api/engine.pyc,sha256=kKAocBMYHIoulIZgTaK7QZtDcnQ2KQqeDf2fILm1LYk,72967
+codemindai_api/pool.pyc,sha256=wPiuq5A8_KzNTyZhcsdMRM0rK9BjAn_Bh0av8dtABuA,52835
+codemindai_api/server.pyc,sha256=54xsYXOuHGHkkIb45mEKC9hSxHOBdb_yKdjPKP0b4QU,116464
+codemindai_api/store.pyc,sha256=AR4S5n_xWPZU8tYJhS277B4CSAF1tqXdCWBxxCzGTpg,58260
+codemindai_api-1.0.0.dist-info/METADATA,sha256=mfVfOxXwK4rTUNCZn5ruBxJKfFq6NM9hRo6m_aXne0Q,8301
+codemindai_api-1.0.0.dist-info/WHEEL,sha256=z11AgxUlkxNJaKutJ-qddoF_F5IpnLT_P57QJW2YQQM,93
+codemindai_api-1.0.0.dist-info/entry_points.txt,sha256=ZRXP1LQNSmJ7_MsOaBXdlE9srPUsaKbLK-RiXp6puFY,59
+codemindai_api-1.0.0.dist-info/top_level.txt,sha256=XENHKjE7GGmBG6MCyL4Odbjb8ZM-3No9qO31iqW9yjI,15
+codemindai_api-1.0.0.dist-info/RECORD,,

codemindai_api-1.0.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: codemindai-api-build
+Root-Is-Purelib: true
+Tag: cp313-none-any

codemindai_api-1.0.0.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+codemindai-api = codemindai_api.cli:main

codemindai_api-1.0.0.dist-info/top_level.txt

@@ -0,0 +1 @@
+codemindai_api