Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77214.70298700381,
-      "required_pages_per_second_upper_estimate": 0.00534189231543567
+      "required_token_positions_per_second": 77044.08675106625,
+      "required_pages_per_second_upper_estimate": 0.0053300886883493
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9651.837873375476,
-      "required_pages_per_second_upper_estimate": 0.004041022964288127
+      "required_token_positions_per_second": 9630.510843883281,
+      "required_pages_per_second_upper_estimate": 0.004032093782398764
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19303.675746750952,
-      "required_pages_per_second_upper_estimate": 0.004020446585885161
+      "required_token_positions_per_second": 19261.021687766563,
+      "required_pages_per_second_upper_estimate": 0.004011562870261892
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28955.51362012643,
-      "required_pages_per_second_upper_estimate": 0.0017358388290172689
+      "required_token_positions_per_second": 28891.532531649846,
+      "required_pages_per_second_upper_estimate": 0.0017320032604565632
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28955.51362012643,
-      "required_pages_per_second_upper_estimate": 0.0022511638221204295
+      "required_token_positions_per_second": 28891.532531649846,
+      "required_pages_per_second_upper_estimate": 0.0022461895739144416
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28955.51362012643,
-      "required_pages_per_second_upper_estimate": 0.3714102387107198
+      "required_token_positions_per_second": 28891.532531649846,
+      "required_pages_per_second_upper_estimate": 0.3705895580052827
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07053940799219305,
-    "producer_p95_seconds": 0.14792634400009774,
-    "gpu_update_seconds": 0.9548440536306839,
-    "producer_batches_per_second": 14.176472818012238,
-    "gpu_batches_per_second": 1.0472914359131378,
-    "producer_headroom": 13.53632077173601,
+    "producer_batch_seconds": 0.06302633353907794,
+    "producer_p95_seconds": 0.14553231399986544,
+    "gpu_update_seconds": 0.9569585818859698,
+    "producer_batches_per_second": 15.866383840652485,
+    "gpu_batches_per_second": 1.044977305108863,
+    "producer_headroom": 15.18347217980927,
     "sustainable": true,
-    "buffer_seconds": 60.155175378733084,
+    "buffer_seconds": 60.2883906588161,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458714.96968259953,
+    "generation_bytes_per_second": 498020.3990320167,
     "upload_bytes_per_second": 5273739.379642944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.42955415590752,
+    "estimated_full_shards_per_hour": 106.86358431072593,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 527300, "loss": 0.5715628266334534, "lr": 0.0003, "grad_norm": 0.15710044105245538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77178990457, "tokens_per_second": 151713.5581075037, "data_wait_s": 1.5919000361463986e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 527400, "loss": 0.5596387386322021, "lr": 0.0003, "grad_norm": 0.17670625021245587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77193611968, "tokens_per_second": 152058.37194807225, "data_wait_s": 1.6325000160577474e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 527500, "loss": 0.601764976978302, "lr": 0.0003, "grad_norm": 0.17139736876850892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77208299186, "tokens_per_second": 152770.9531720846, "data_wait_s": 1.522999991721008e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 527600, "loss": 0.563929557800293, "lr": 0.0003, "grad_norm": 0.15762018683778434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77223013318, "tokens_per_second": 153284.73274663766, "data_wait_s": 1.0989000202243915e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5ff6265cb45151eb2ad9666e92a5795ce11720ec4a591958eafe8e228e77c62
-size 53115
+oid sha256:7f6e84f822280fa21642df4a7a2f44a642289936e2e65a691452ae5670d18ce9
+size 55233