idemcore-verify 0.3.0__py3-none-any.whl

idemcore_verify/__init__.py

@@ -0,0 +1,5 @@
+"""idemcore-verify: check Idemcore per-token hash logs.
+
+Pure Python standard library. No model, no kernel, no GPU needed.
+"""
+__version__ = "0.3.0"

idemcore_verify/__main__.py

@@ -0,0 +1,121 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Command line: python -m idemcore_verify {check,compare,check-batch,compare-batch,samples} ..."""
+from __future__ import annotations
+
+import argparse
+import dataclasses
+import json
+import os
+import sys
+
+from .core import (check_batch_log, check_log, compare_batch, compare_logs,
+                   parse_batch_log, parse_log)
+from . import __version__
+from .samples import run_samples
+
+
+def _print_check(path, res, as_json):
+    if as_json:
+        print(json.dumps({"file": path, **dataclasses.asdict(res)}))
+        return
+    status = "OK  " if res.ok else "FAIL"
+    print(f"{status} {path}: {res.n_tokens} tokens, ALL {res.all_claimed} "
+          f"(recomputed {res.all_recomputed})")
+    for e in res.errors[:20]:
+        print(f"     - {e}")
+
+
+def main(argv=None) -> int:
+    ap = argparse.ArgumentParser(prog="idemcore-verify",
+                                 description="Check Idemcore per-token hash logs.")
+    ap.add_argument("--version", action="version", version=f"%(prog)s {__version__}")
+    sub = ap.add_subparsers(dest="cmd", required=True)
+    c = sub.add_parser("check", help="validate a log and recompute its ALL chain")
+    c.add_argument("logs", nargs="+")
+    c.add_argument("--json", action="store_true")
+    m = sub.add_parser("compare", help="compare a candidate log against a reference log")
+    m.add_argument("reference")
+    m.add_argument("candidate")
+    m.add_argument("--subset", action="store_true",
+                   help="allow one log to cover only part of the other (spot check)")
+    m.add_argument("--json", action="store_true")
+    cb = sub.add_parser("check-batch", help="validate a batched (multi-stream) log and recompute every stream's ALL")
+    cb.add_argument("logs", nargs="+")
+    mb = sub.add_parser("compare-batch",
+                        help="compare every row of a batched log against a single-stream reference log")
+    mb.add_argument("reference")
+    mb.add_argument("batch")
+    sm = sub.add_parser("samples", help="one-command self-check: recompute every published hash in "
+                        "samples/HASHES.tsv and re-check every pair in samples/PAIRS.tsv")
+    sm.add_argument("--dir", default=None, help="samples directory (default: ./samples)")
+    sm.add_argument("-v", "--verbose", action="store_true", help="print one line per file and pair")
+    a = ap.parse_args(argv)
+
+    if a.cmd == "samples":
+        d = a.dir or ("samples" if os.path.isdir("samples") else
+                      os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "samples"))
+        if not os.path.isfile(os.path.join(d, "HASHES.tsv")):
+            print(f"idemcore-verify: no samples found at {os.path.abspath(d)}\n"
+                  "The sample logs are not part of the pip package. Run this command inside a\n"
+                  "clone of https://github.com/idemcore/idemcore-verifier or the unzipped release\n"
+                  "archive, or pass --dir PATH/TO/samples.", file=sys.stderr)
+            return 2
+        return run_samples(d, verbose=a.verbose)
+
+    if a.cmd == "check-batch":
+        ok = True
+        for p in a.logs:
+            r = check_batch_log(parse_batch_log(p))
+            print(f"{'OK  ' if r.ok else 'FAIL'} {p}: {r.n_streams} streams, {r.n_rows} rows "
+                  f"({r.n_filler_rows} filler), ALL[0] {r.all_by_stream.get(0)}, STREAMS {r.streams_digest}")
+            for e in r.errors[:20]:
+                print(f"     - {e}")
+            ok &= r.ok
+        return 0 if ok else 1
+
+    if a.cmd == "compare-batch":
+        ref, b = parse_log(a.reference), parse_batch_log(a.batch)
+        rr, rb = check_log(ref), check_batch_log(b)
+        res = compare_batch(ref, b)
+        ok = rr.ok and rb.ok and res.ok
+        _print_check(a.reference, rr, False)
+        print(f"{'OK  ' if rb.ok else 'FAIL'} {a.batch}: {rb.n_streams} streams, {rb.n_rows} rows")
+        for e in rb.errors[:20]:
+            print(f"     - {e}")
+        print(f"{'MATCH' if ok else 'DIVERGE'}: {res.matched}/{res.compared} batched rows identical to the "
+              f"reference (missing in reference {res.missing_in_ref})")
+        if res.first_divergence:
+            print(f"  first divergence: {res.first_divergence}")
+        return 0 if ok else 1
+
+    if a.cmd == "check":
+        ok = True
+        for p in a.logs:
+            r = check_log(parse_log(p))
+            _print_check(p, r, a.json)
+            ok &= r.ok
+        return 0 if ok else 1
+
+    ref, cand = parse_log(a.reference), parse_log(a.candidate)
+    rr, rc = check_log(ref), check_log(cand)
+    res = compare_logs(ref, cand, subset=a.subset)
+    ok = rr.ok and rc.ok and res.ok
+    if a.json:
+        print(json.dumps({"ok": ok, "reference": dataclasses.asdict(rr),
+                          "candidate": dataclasses.asdict(rc),
+                          "compare": dataclasses.asdict(res)}))
+    else:
+        _print_check(a.reference, rr, False)
+        _print_check(a.candidate, rc, False)
+        print(f"{'MATCH' if ok else 'DIVERGE'}: {res.matched}/{res.compared} positions identical "
+              f"(mode {res.mode}; only-in-ref {res.only_in_ref}, only-in-cand {res.only_in_cand})")
+        if res.first_divergence:
+            d = res.first_divergence
+            print(f"  first divergence at payload {d['pi']} position {d['j']}: fields {d['fields']}")
+            print(f"    ref  (line {d['ref_line']}): {d['ref']}")
+            print(f"    cand (line {d['cand_line']}): {d['cand']}")
+    return 0 if ok else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())

idemcore_verify/core.py

@@ -0,0 +1,357 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Parsing and checking of Idemcore per-token hash logs (log format v1).
+
+A log is a text file with one line per generated/scored token:
+
+    <pi> <j> <h> <hc> <freq> <total>
+
+  pi     payload (sequence) index, decimal
+  j      token position inside the payload, decimal
+  h      FNV-1a-64 of the V int32 logits (little-endian), 16 hex digits
+  hc     FNV-1a-64 of the (V+1) uint64 integer CDF table (little-endian), 16 hex
+  freq   CDF width of the observed token: CUM[t+1] - CUM[t], decimal
+  total  CDF total CUM[V], decimal, 0 < total < 2**31
+
+The hosted API serves the *public* form of the same log
+(``idemcore-log/r1-public``): the first four fields only,
+
+    <pi> <j> <h> <hc>
+
+This verifier accepts both forms. ``ALL`` depends only on ``h`` and ``hc``, so
+it is the same for the full and the public form of a run.
+
+followed by a chain line
+
+    ALL <16 hex>
+
+where ALL = fold over all token lines, in file order, of
+    a = fnv1a64(le64(h), a); a = fnv1a64(le64(hc), a)
+starting from the FNV-1a-64 offset basis.  Other lines that start with a
+keyword (not a number) are reserved for future versions and ignored.
+"""
+from __future__ import annotations
+
+import dataclasses
+import gzip
+from typing import Dict, Iterable, List, Optional, Tuple
+
+FNV_OFFSET = 0xCBF29CE484222325
+FNV_PRIME = 0x100000001B3
+MASK64 = (1 << 64) - 1
+
+
+def fnv1a64(data: bytes, h: int = FNV_OFFSET) -> int:
+    for b in data:
+        h ^= b
+        h = (h * FNV_PRIME) & MASK64
+    return h
+
+
+def fnv1a64_u64(v: int, h: int) -> int:
+    """FNV-1a-64 over the 8 little-endian bytes of v (no allocation)."""
+    for i in range(8):
+        h ^= (v >> (8 * i)) & 0xFF
+        h = (h * FNV_PRIME) & MASK64
+    return h
+
+
+@dataclasses.dataclass(frozen=True)
+class TokenLine:
+    pi: int
+    j: int
+    h: int
+    hc: int
+    freq: Optional[int]   # None in a public (4-field) log line
+    total: Optional[int]
+    lineno: int
+
+    @property
+    def key(self) -> Tuple[int, int]:
+        return (self.pi, self.j)
+
+
+@dataclasses.dataclass
+class Log:
+    path: str
+    tokens: List[TokenLine]
+    all_hex: Optional[str]
+    errors: List[str]
+
+    def chain(self) -> int:
+        a = FNV_OFFSET
+        for t in self.tokens:
+            a = fnv1a64_u64(t.h, a)
+            a = fnv1a64_u64(t.hc, a)
+        return a
+
+    def by_key(self) -> Dict[Tuple[int, int], TokenLine]:
+        return {t.key: t for t in self.tokens}
+
+
+def read_lines(path: str) -> List[str]:
+    """Read a log as ASCII lines; files ending in .gz are decompressed."""
+    opener = gzip.open if path.endswith(".gz") else open
+    with opener(path, "rt", encoding="ascii") as f:
+        return f.read().splitlines()
+
+
+def parse_log(path: str, lines: Optional[Iterable[str]] = None) -> Log:
+    if lines is None:
+        lines = read_lines(path)
+    toks: List[TokenLine] = []
+    errors: List[str] = []
+    all_hex = None
+    seen = set()
+    for n, raw in enumerate(lines, 1):
+        s = raw.strip()
+        if not s or s.startswith("#"):
+            continue
+        p = s.split()
+        if p[0] == "ALL":
+            if len(p) != 2 or len(p[1]) != 16:
+                errors.append(f"line {n}: malformed ALL line")
+            all_hex = p[1].lower()
+            continue
+        if not p[0].isdigit():
+            continue  # reserved keyword line (format v1 ignores it)
+        if len(p) not in (4, 6):
+            errors.append(f"line {n}: expected 6 fields (or 4 in a public log), got {len(p)}")
+            continue
+        try:
+            pi, j = int(p[0]), int(p[1])
+            if len(p[2]) != 16 or len(p[3]) != 16:
+                raise ValueError("hash field is not 16 hex digits")
+            h, hc = int(p[2], 16), int(p[3], 16)
+            freq = total = None
+            if len(p) == 6:
+                freq, total = int(p[4]), int(p[5])
+        except ValueError as e:
+            errors.append(f"line {n}: {e}")
+            continue
+        if total is not None and not (0 < total < 2 ** 31):
+            errors.append(f"line {n}: total {total} out of range (0, 2^31)")
+        if freq is not None and total is not None and not (0 < freq <= total):
+            errors.append(f"line {n}: freq {freq} not in (0, total]")
+        if (pi, j) in seen:
+            errors.append(f"line {n}: duplicate position ({pi},{j})")
+        seen.add((pi, j))
+        toks.append(TokenLine(pi, j, h, hc, freq, total, n))
+    if all_hex is None:
+        errors.append("no ALL line")
+    return Log(path, toks, all_hex, errors)
+
+
+@dataclasses.dataclass
+class CheckResult:
+    ok: bool
+    n_tokens: int
+    all_claimed: Optional[str]
+    all_recomputed: str
+    errors: List[str]
+
+
+def check_log(log: Log) -> CheckResult:
+    rec = "%016x" % log.chain()
+    errs = list(log.errors)
+    if log.all_hex is not None and rec != log.all_hex:
+        errs.append(f"ALL mismatch: file says {log.all_hex}, recomputed {rec}")
+    return CheckResult(not errs, len(log.tokens), log.all_hex, rec, errs)
+
+
+@dataclasses.dataclass
+class CompareResult:
+    ok: bool
+    compared: int
+    matched: int
+    only_in_ref: int
+    only_in_cand: int
+    first_divergence: Optional[dict]
+    mode: str
+
+
+def _diff_fields(r: TokenLine, c: TokenLine) -> List[str]:
+    """Fields that differ. freq/total are compared only when both lines carry
+    them (a public 4-field log line has neither)."""
+    out = [f for f in ("h", "hc") if getattr(r, f) != getattr(c, f)]
+    if r.freq is not None and c.freq is not None:
+        out += [f for f in ("freq", "total") if getattr(r, f) != getattr(c, f)]
+    return out
+
+
+def compare_logs(ref: Log, cand: Log, subset: bool = False) -> CompareResult:
+    """Compare candidate against reference position by position.
+
+    subset=False: both logs must cover exactly the same positions.
+    subset=True : every position in the *smaller* log must appear, identical,
+                  in the larger one (e.g. a CPU spot-check of a GPU run).
+    """
+    R, C = ref.by_key(), cand.by_key()
+    keys = sorted(set(R) & set(C))
+    matched = 0
+    first = None
+    for k in keys:
+        r, c = R[k], C[k]
+        diff = _diff_fields(r, c)
+        if not diff:
+            matched += 1
+        elif first is None:
+            first = {
+                "pi": k[0], "j": k[1], "fields": diff,
+                "ref": {f: (("%016x" % getattr(r, f)) if f in ("h", "hc") else getattr(r, f)) for f in diff},
+                "cand": {f: (("%016x" % getattr(c, f)) if f in ("h", "hc") else getattr(c, f)) for f in diff},
+                "ref_line": r.lineno, "cand_line": c.lineno,
+            }
+    only_r = len(set(R) - set(C))
+    only_c = len(set(C) - set(R))
+    if subset:
+        coverage_ok = min(only_r, only_c) == 0
+    else:
+        coverage_ok = only_r == 0 and only_c == 0
+    ok = coverage_ok and matched == len(keys) and len(keys) > 0
+    return CompareResult(ok, len(keys), matched, only_r, only_c, first,
+                         "subset" if subset else "exact")
+
+
+# ---------------------------------------------------------------------------
+# Batched (multi-stream) logs, format v1-batch.  See SPEC.md, "Batched logs".
+#
+#     <s> <pi> <j> <h> <hc> <freq> <total>      one line per token of stream s
+#     ALL <s> <16 hex>                          chain of stream s, in file order
+#
+# Stream ids >= FILLER_BASE mark filler work that occupied a slot before a
+# staggered stream joined.  Filler rows have no ALL line; they are compared
+# against the reference like every other row.
+
+FILLER_BASE = 1000
+
+
+@dataclasses.dataclass
+class BatchLog:
+    path: str
+    rows: List[Tuple[int, TokenLine]]          # (stream, token line), file order
+    all_by_stream: Dict[int, str]
+    errors: List[str]
+
+    def streams(self) -> List[int]:
+        return sorted({s for s, _ in self.rows if s < FILLER_BASE})
+
+    def chains(self) -> Dict[int, int]:
+        acc: Dict[int, int] = {}
+        for s, t in self.rows:
+            if s >= FILLER_BASE:
+                continue
+            a = acc.get(s, FNV_OFFSET)
+            a = fnv1a64_u64(t.h, a)
+            acc[s] = fnv1a64_u64(t.hc, a)
+        return acc
+
+
+def parse_batch_log(path: str, lines: Optional[Iterable[str]] = None) -> BatchLog:
+    if lines is None:
+        lines = read_lines(path)
+    rows: List[Tuple[int, TokenLine]] = []
+    alls: Dict[int, str] = {}
+    errors: List[str] = []
+    for n, raw in enumerate(lines, 1):
+        s = raw.strip()
+        if not s or s.startswith("#"):
+            continue
+        p = s.split()
+        if p[0] == "ALL":
+            if len(p) != 3 or not p[1].isdigit() or len(p[2]) != 16:
+                errors.append(f"line {n}: malformed batch ALL line")
+                continue
+            sid = int(p[1])
+            if sid in alls:
+                errors.append(f"line {n}: second ALL line for stream {sid}")
+            alls[sid] = p[2].lower()
+            continue
+        if not p[0].isdigit():
+            continue  # reserved keyword line
+        if len(p) != 7:
+            errors.append(f"line {n}: expected 7 fields, got {len(p)}")
+            continue
+        try:
+            sid, pi, j = int(p[0]), int(p[1]), int(p[2])
+            if len(p[3]) != 16 or len(p[4]) != 16:
+                raise ValueError("hash field is not 16 hex digits")
+            h, hc = int(p[3], 16), int(p[4], 16)
+            freq, total = int(p[5]), int(p[6])
+        except ValueError as e:
+            errors.append(f"line {n}: {e}")
+            continue
+        if not (0 < total < 2 ** 31):
+            errors.append(f"line {n}: total {total} out of range (0, 2^31)")
+        if not (0 < freq <= total):
+            errors.append(f"line {n}: freq {freq} not in (0, total]")
+        rows.append((sid, TokenLine(pi, j, h, hc, freq, total, n)))
+    return BatchLog(path, rows, alls, errors)
+
+
+def streams_digest(all_by_stream: Dict[int, str]) -> str:
+    """One 64-bit summary of a batched log: fold over streams in ascending id of
+    le64(stream id) then le64(ALL of that stream)."""
+    a = FNV_OFFSET
+    for sid in sorted(all_by_stream):
+        a = fnv1a64_u64(sid, a)
+        a = fnv1a64_u64(int(all_by_stream[sid], 16), a)
+    return "%016x" % a
+
+
+@dataclasses.dataclass
+class BatchCheckResult:
+    ok: bool
+    n_streams: int
+    n_rows: int
+    n_filler_rows: int
+    all_by_stream: Dict[int, str]
+    streams_digest: str
+    errors: List[str]
+
+
+def check_batch_log(log: BatchLog) -> BatchCheckResult:
+    errs = list(log.errors)
+    rec = {s: "%016x" % v for s, v in log.chains().items()}
+    for s in sorted(set(rec) | set(log.all_by_stream)):
+        if s not in log.all_by_stream:
+            errs.append(f"stream {s}: no ALL line")
+        elif s not in rec:
+            errs.append(f"stream {s}: ALL line but no token lines")
+        elif rec[s] != log.all_by_stream[s]:
+            errs.append(f"stream {s}: ALL mismatch: file says {log.all_by_stream[s]}, recomputed {rec[s]}")
+    filler = sum(1 for s, _ in log.rows if s >= FILLER_BASE)
+    return BatchCheckResult(not errs, len(rec), len(log.rows), filler, rec,
+                            streams_digest(rec), errs)
+
+
+@dataclasses.dataclass
+class BatchCompareResult:
+    ok: bool
+    compared: int
+    matched: int
+    missing_in_ref: int
+    first_divergence: Optional[dict]
+
+
+def compare_batch(ref: Log, batch: BatchLog) -> BatchCompareResult:
+    """Every row of every stream (filler included) must equal the reference
+    token line at the same (payload, position)."""
+    R = ref.by_key()
+    matched = missing = 0
+    first = None
+    for sid, t in batch.rows:
+        r = R.get(t.key)
+        if r is None:
+            missing += 1
+            if first is None:
+                first = {"stream": sid, "pi": t.pi, "j": t.j, "fields": ["missing in reference"],
+                         "cand_line": t.lineno}
+            continue
+        diff = _diff_fields(r, t)
+        if not diff:
+            matched += 1
+        elif first is None:
+            first = {"stream": sid, "pi": t.pi, "j": t.j, "fields": diff,
+                     "ref_line": r.lineno, "cand_line": t.lineno}
+    ok = missing == 0 and matched == len(batch.rows) and matched > 0
+    return BatchCompareResult(ok, len(batch.rows), matched, missing, first)

idemcore_verify/samples.py

@@ -0,0 +1,92 @@
+# SPDX-License-Identifier: Apache-2.0
+"""One-command self-check of the published evidence.
+
+1. For every row of samples/HASHES.tsv, recompute the log's digest(s) from its
+   per-token lines and compare with the published value.
+2. For every row of samples/PAIRS.tsv, compare candidate against reference
+   (modes: exact, subset, batch).
+Prints PASS or FAIL and returns 0 / 1.
+"""
+from __future__ import annotations
+
+import os
+from typing import List
+
+from .core import (check_batch_log, check_log, compare_batch, compare_logs,
+                   parse_batch_log, parse_log)
+
+
+def _rows(path: str) -> List[List[str]]:
+    out = []
+    with open(path, encoding="utf-8") as f:
+        for line in f:
+            line = line.rstrip("\n")
+            if not line.strip() or line.startswith("#"):
+                continue
+            out.append(line.split("\t"))
+    return out
+
+
+def run_samples(d: str, verbose: bool = False) -> int:
+    failures: List[str] = []
+    n_files = n_pairs = n_rows = 0
+    cache = {}
+
+    def single(rel):
+        if rel not in cache:
+            cache[rel] = parse_log(os.path.join(d, rel))
+        return cache[rel]
+
+    print(f"idemcore-verify self-check of {os.path.abspath(d)}")
+    print("[1/2] recompute published hashes (HASHES.tsv)")
+    for r in _rows(os.path.join(d, "HASHES.tsv")):
+        rel, kind, exp_all, exp_streams = r[0], r[1], r[8], r[9]
+        n_files += 1
+        try:
+            if kind == "single":
+                c = check_log(single(rel))
+                got, got_s = c.all_recomputed, "-"
+                ok = c.ok and got == exp_all
+                n_rows += c.n_tokens
+            else:
+                c = check_batch_log(parse_batch_log(os.path.join(d, rel)))
+                got, got_s = c.all_by_stream.get(0, "-"), c.streams_digest
+                ok = c.ok and got == exp_all and got_s == exp_streams
+                n_rows += c.n_rows
+        except OSError as e:
+            ok, got, got_s = False, f"unreadable ({e})", "-"
+        if not ok:
+            failures.append(f"hash  {rel}: published {exp_all}/{exp_streams}, recomputed {got}/{got_s}")
+        if verbose or not ok:
+            print(f"  {'PASS' if ok else 'FAIL'} {rel}  ALL {got}" + ("" if got_s == "-" else f"  STREAMS {got_s}"))
+
+    print("[2/2] re-check reference/candidate pairs (PAIRS.tsv)")
+    for r in _rows(os.path.join(d, "PAIRS.tsv")):
+        ref, cand, mode, what = r[0], r[1], r[2], r[3] if len(r) > 3 else ""
+        n_pairs += 1
+        try:
+            if mode == "batch":
+                b = parse_batch_log(os.path.join(d, cand))
+                res = compare_batch(single(ref), b)
+                ok = res.ok and check_batch_log(b).ok and check_log(single(ref)).ok
+                detail = f"{res.matched}/{res.compared} rows"
+            else:
+                rl, cl = single(ref), single(cand)
+                res = compare_logs(rl, cl, subset=(mode == "subset"))
+                ok = res.ok and check_log(rl).ok and check_log(cl).ok
+                detail = f"{res.matched}/{res.compared} positions"
+        except OSError as e:
+            ok, detail = False, f"unreadable ({e})"
+        if not ok:
+            failures.append(f"pair  {what}: {detail}")
+        if verbose or not ok:
+            print(f"  {'PASS' if ok else 'FAIL'} {what}  ({detail})")
+
+    print(f"checked {n_files} logs ({n_rows} token rows) and {n_pairs} pairs")
+    if failures:
+        print(f"FAIL: {len(failures)} problem(s)")
+        for f in failures:
+            print("  - " + f)
+        return 1
+    print("PASS: every published hash recomputes and every pair is bit-identical")
+    return 0

idemcore_verify-0.3.0.dist-info/METADATA

@@ -0,0 +1,223 @@
+Metadata-Version: 2.4
+Name: idemcore-verify
+Version: 0.3.0
+Summary: Check Idemcore per-token hash logs: verify, without trusting us, that two machines produced bit-identical LLM inference.
+Author-email: Don Wick <don.wick@idemcore.net>
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://idemcore.net
+Project-URL: Paper, https://doi.org/10.5281/zenodo.23851129
+Project-URL: Source, https://github.com/idemcore/idemcore-verifier
+Project-URL: Issues, https://github.com/idemcore/idemcore-verifier/issues
+Keywords: reproducibility,deterministic inference,llm,verification,integer inference
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Testing
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Dynamic: license-file
+
+# idemcore-verifier
+
+[![verify](https://github.com/idemcore/idemcore-verifier/actions/workflows/verify.yml/badge.svg?branch=main)](https://github.com/idemcore/idemcore-verifier/actions/workflows/verify.yml)
+
+**Check, without trusting us, that two machines produced bit-identical LLM inference.**
+
+Project site: <https://idemcore.net>
+
+Idemcore is an integer-only inference engine: the same model, prompt, and
+settings give the *same bits* on an x86 laptop, an ARM phone, and NVIDIA GPUs
+from Turing to Hopper. Every token it scores or generates emits a 64-bit hash
+of the full logit vector and a 64-bit hash of the integer sampling table
+(CDF). A run log is a list of these hashes plus one chained digest (`ALL`).
+
+This repository is the **open half** of Idemcore:
+
+* the log format and hash protocol ([SPEC.md](SPEC.md)),
+* a small, dependency-free Python verifier (`idemcore_verify/`),
+* the measurement methodology ([METHODOLOGY.md](METHODOLOGY.md)),
+* raw per-token logs from real cross-device runs (`samples/`) and the results
+  table they back ([RESULTS.md](RESULTS.md)).
+
+The inference kernel itself is not in this repository and is not open source.
+You do not need it to check a claim: the verifier only needs two logs.
+
+## Reproduce in 5 minutes
+
+You need `git` and Python ≥ 3.9. No GPU, no model and no third-party package.
+
+```sh
+# 1. get the verifier and the sample logs (about 6 MB)
+git clone https://github.com/idemcore/idemcore-verifier.git
+cd idemcore-verifier
+
+# 2. optional: a clean virtual environment and the `idemcore-verify` command
+python3 -m venv .venv && . .venv/bin/activate
+pip install .
+
+# 3. recompute every published hash and re-check every published pair (~5 s)
+python3 -m idemcore_verify samples        # same as: make verify  or  sh verify_samples.sh
+
+# 4. check one pair yourself: x86-64 CPU vs NVIDIA L40S, gpt-oss-20b
+python3 -m idemcore_verify compare samples/gpt-oss-20b/x86_64.log \
+                                   samples/gpt-oss-20b/nvidia_l40s.log
+
+# 5. the unit tests, including a tamper test
+python3 -m unittest discover -s tests
+```
+
+Expected output of step 3:
+
+```
+checked 65 logs (448904 token rows) and 58 pairs
+PASS: every published hash recomputes and every pair is bit-identical
+```
+
+and of step 4:
+
+```
+OK   samples/gpt-oss-20b/x86_64.log: 512 tokens, ALL 8ec348aaf3b4109c (recomputed 8ec348aaf3b4109c)
+OK   samples/gpt-oss-20b/nvidia_l40s.log: 512 tokens, ALL 8ec348aaf3b4109c (recomputed 8ec348aaf3b4109c)
+MATCH: 512/512 positions identical (mode exact; only-in-ref 0, only-in-cand 0)
+```
+
+Step 3 recomputes all 65 published digests in
+[samples/HASHES.tsv](samples/HASHES.tsv) from the shipped logs (about 450 000
+token rows, including batched GPU logs) and re-checks all 58 pairs in
+[samples/PAIRS.tsv](samples/PAIRS.tsv). To prove the check is not a rubber
+stamp, edit any hex digit in a log and run it again: the chain breaks.
+
+`idemcore-verify samples` looks for `./samples` in the current directory; run
+it from the repository root, or pass `--dir path/to/samples`. `make` is not
+required; every `make` target is a one-line Python command.
+
+### Check a live answer from api.idemcore.net
+
+The hosted API (`https://api.idemcore.net/v1`, OpenAI-compatible) attaches a
+receipt to every answer, and `GET /v1/receipts/{id}/log` returns that answer's
+per-token log. The API serves the public form of the log (`log_format`
+`idemcore-log/r1-public`): the `pi j h hc` fields of [SPEC.md](SPEC.md) plus the
+`ALL` line, without `freq` and `total`. The verifier accepts both forms (version
+0.3.0 and later). Access needs an API key: sign in to the console at
+<https://console.idemcore.net/#/login> and create one. With a key, save the log and run:
+
+```sh
+curl -s https://api.idemcore.net/v1/receipts/RECEIPT_ID/log \
+  -H "Authorization: Bearer $IDEMCORE_API_KEY" | jq -r .log_text > answer.log
+python3 -m idemcore_verify check answer.log
+python3 -m idemcore_verify compare answer.log your_rerun.log
+```
+
+### Install as a package
+
+The verifier is published on PyPI as
+[`idemcore-verify`](https://pypi.org/project/idemcore-verify/) (console script of
+the same name):
+
+```sh
+pip install idemcore-verify
+idemcore-verify --version
+idemcore-verify check run.log
+idemcore-verify compare reference.log candidate.log
+```
+
+The package contains only the verifier; the sample logs ship with this
+repository.
+
+All hash tables, per device, workload and batch size: [EVIDENCE.md](EVIDENCE.md).
+
+Requires Python ≥ 3.9, standard library only.
+
+## What the verifier checks
+
+1. **Well-formedness** – every line has six fields, hashes are 64-bit hex,
+   `0 < freq ≤ total < 2^31`, no duplicate positions.
+2. **Chain integrity** – it recomputes the `ALL` digest from the per-token
+   lines; editing a single line breaks it.
+3. **Agreement** – it compares a candidate log against a reference log
+   position by position, keyed by `(payload, position)`, and prints the first
+   divergence (which token, which field). `--subset` allows a short spot check
+   (e.g. 128 CPU tokens against a 512-token GPU run).
+
+A match on `h` means the two machines computed the **same integer logits for
+the whole vocabulary** at that position, not just the same argmax token. A
+match on `hc` means they built the same integer sampling distribution, so any
+seeded sampler driven by it makes the same choices.
+
+## Published pairs
+
+| Model | Reference | Candidate | Positions | Result |
+|---|---|---|---|---|
+| Qwen2.5-0.5B, 8-bit | x86-64 CPU | NVIDIA T4 | 3260 | identical |
+| Qwen2.5-0.5B, 8-bit | x86-64 CPU | Galaxy Z Fold5 (Android) | 240 | identical |
+| Qwen2.5-0.5B, 8-bit | x86-64 CPU | Windows 11 ARM64 laptop (Snapdragon X Plus) | 240 | identical |
+| Qwen2.5-7B, 8-bit | x86-64 CPU | NVIDIA A10 | 512 | identical |
+| Qwen2.5-7B, 8-bit | x86-64 CPU | Galaxy Z Fold5 (Android) | 64 (prefix) | identical |
+| Qwen2.5-7B, 4-bit (g) | x86-64 CPU | NVIDIA A10 | 512 | identical |
+| Qwen2.5-7B, 4-bit (g4) | x86-64 CPU | NVIDIA A10G | 512 | identical |
+| Qwen2.5-7B, 4-bit (g) | x86-64 CPU | Galaxy Z Fold5 (Android), both CPU code paths | 128 (prefix) | identical |
+| Qwen2.5-7B, 4-bit (g) | x86-64 CPU | x86-64 CPU, 6 repeat runs (6 and 8 pinned threads) | 512 | identical |
+| Qwen2.5-7B, 4-bit (g) | x86-64 CPU | NVIDIA A10G and A10, batched B = 1–64, shuffled and staggered (40 logs) | every row of every stream | identical |
+| gpt-oss-20b (MoE) | x86-64 CPU | NVIDIA L40S | 512 | identical |
+| gpt-oss-120b (MoE) | x86-64 CPU (spot check) | NVIDIA H100 | 128 of 512 | identical |
+
+Details per file: [samples/MANIFEST.md](samples/MANIFEST.md).
+Qwen2.5-7B 4-bit on Android runs at 3.39 tok/s on the Fold5 (see [RESULTS.md](RESULTS.md)).
+
+**Batched GPU serving.** With an improved batched GPU kernel, Qwen2.5-7B 4-bit
+on an NVIDIA A10G serves up to 64 concurrent streams, and every stream's
+per-token hashes match a single-stream x86-64 run at batch sizes 1, 8, 16, 32
+and 64, including shuffled slot order and staggered joins (0 mismatches).
+The same holds on an NVIDIA A10. Aggregate throughput (A10G): 641 tok/s at 8
+streams and 1372 tok/s at 32. Numbers and caveats:
+[RESULTS.md](RESULTS.md#2b-batched-serving-qwen25-7b-4-bit-g-nvidia-a10g).
+The raw batched logs are in `samples/batched/` and are re-checked by step 3 above.
+
+## Methodology and benchmarks
+
+* How every number was measured (workloads, clocks, baselines, energy,
+  quality): [METHODOLOGY.md](METHODOLOGY.md).
+* Results and their caveats: [RESULTS.md](RESULTS.md); every hash, per device,
+  workload and batch size: [EVIDENCE.md](EVIDENCE.md).
+* The serving-benchmark kit used in the paper will be published separately and
+  linked here.
+
+## Paper
+
+Don Wick, *Idemcore: Bit-Identical LLM Inference from Phones to Datacenter
+GPUs, with an Open Per-Token Verification Protocol*, preprint, Zenodo,
+2026. DOI: [10.5281/zenodo.23851129](https://doi.org/10.5281/zenodo.23851129).
+Citation metadata: [CITATION.cff](CITATION.cff).
+
+## What this does **not** prove
+
+* The logs are produced by our engine. A matching pair shows two runs agree;
+  it does not prove *which* program produced them. To check a provider, re-run
+  the same request on hardware you control (or ask a second party to) and
+  compare logs.
+* The hash is FNV-1a-64: fast and good for detecting accidental divergence,
+  **not** collision-resistant against an adversary who can choose logits. A
+  SHA-256 / BLAKE3 variant of the log is planned; the format has a version.
+
+## License
+
+Apache-2.0 for everything in this repository (verifier, spec, methodology,
+sample logs). See [LICENSE](LICENSE). Idemcore's inference kernel, weight
+format, and tables are proprietary and are not covered by this license.
+
+Contact: Don Wick — don.wick@idemcore.net · Website: <https://idemcore.net>
+· Security reports: [SECURITY.md](SECURITY.md) · Mismatch reports: open an
+issue with the "Hash mismatch" template.
+
+Copyright 2026 Don Wick.

idemcore_verify-0.3.0.dist-info/RECORD

@@ -0,0 +1,10 @@
+idemcore_verify/__init__.py,sha256=gpftuwNdwa7CGtbU6x8Y4_gIOZ6BdL3jTGgvPT6Cv5Q,149
+idemcore_verify/__main__.py,sha256=yZ9Vs65UbIezebT1MCLqtuKsyCydF7XHJ5ip8Au0tJU,5580
+idemcore_verify/core.py,sha256=4cWgASveRoQh9c9bkSDFwBCJUrZnvTSNFnTugqZ8bpk,12175
+idemcore_verify/samples.py,sha256=34Rq334L32sxGM4RPgYsf0gNYrGFLOtg0bKu0pFTk6o,3626
+idemcore_verify-0.3.0.dist-info/licenses/LICENSE,sha256=M7IiqWOs_rWTLS1hbDnGU3e5lhrKG1kyOjMT9MUzNtA,11339
+idemcore_verify-0.3.0.dist-info/METADATA,sha256=1rYg0pPFKDKke64Uj1S6FkKYQlOQkvLBJDlJ_wQ6pIo,10289
+idemcore_verify-0.3.0.dist-info/WHEEL,sha256=mOdXR9C3wMVov7cSY8PY-XCzo3SmmWCHvGQr5csJoao,91
+idemcore_verify-0.3.0.dist-info/entry_points.txt,sha256=8q2hVhhzQ03h95cxDmZsBcItz4o0yOWPbFjL5_Tev6c,66
+idemcore_verify-0.3.0.dist-info/top_level.txt,sha256=cfSZ0Bfeeky3aUekYWnxxnDSt_Ih8cPtWu8Lr7Yti9k,16
+idemcore_verify-0.3.0.dist-info/RECORD,,

idemcore_verify-0.3.0.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (84.2.1)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

idemcore_verify-0.3.0.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+idemcore-verify = idemcore_verify.__main__:main

idemcore_verify-0.3.0.dist-info/licenses/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Don Wick
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

idemcore_verify-0.3.0.dist-info/top_level.txt

@@ -0,0 +1 @@
+idemcore_verify