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": 76817.66972260561,
-      "required_pages_per_second_upper_estimate": 0.005314424632960506
+      "required_token_positions_per_second": 77214.70298700381,
+      "required_pages_per_second_upper_estimate": 0.00534189231543567
     },
     {
       "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": 9602.208715325702,
-      "required_pages_per_second_upper_estimate": 0.004020244272187356
+      "required_token_positions_per_second": 9651.837873375476,
+      "required_pages_per_second_upper_estimate": 0.004041022964288127
     },
     {
       "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": 19204.417430651403,
-      "required_pages_per_second_upper_estimate": 0.003999773696259448
+      "required_token_positions_per_second": 19303.675746750952,
+      "required_pages_per_second_upper_estimate": 0.004020446585885161
     },
     {
       "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": 28806.626145977105,
-      "required_pages_per_second_upper_estimate": 0.0017269132522800266
+      "required_token_positions_per_second": 28955.51362012643,
+      "required_pages_per_second_upper_estimate": 0.0017358388290172689
     },
     {
       "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": 28806.626145977105,
-      "required_pages_per_second_upper_estimate": 0.0022395884758921077
+      "required_token_positions_per_second": 28955.51362012643,
+      "required_pages_per_second_upper_estimate": 0.0022511638221204295
     },
     {
       "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": 28806.626145977105,
-      "required_pages_per_second_upper_estimate": 0.36950047005524694
+      "required_token_positions_per_second": 28955.51362012643,
+      "required_pages_per_second_upper_estimate": 0.3714102387107198
     }
   ],
   "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": 0.06570874442968488,
+    "producer_batch_seconds": 0.07053940799219305,
     "producer_p95_seconds": 0.14792634400009774,
-    "gpu_update_seconds": 0.9597791792726511,
-    "producer_batches_per_second": 15.218674602283762,
-    "gpu_batches_per_second": 1.0419063276178062,
-    "producer_headroom": 14.606567019397449,
+    "gpu_update_seconds": 0.9548440536306839,
+    "producer_batches_per_second": 14.176472818012238,
+    "gpu_batches_per_second": 1.0472914359131378,
+    "producer_headroom": 13.53632077173601,
     "sustainable": true,
-    "buffer_seconds": 60.46608829417702,
+    "buffer_seconds": 60.155175378733084,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 463207.0805600846,
-    "upload_bytes_per_second": 5639213.806782705,
+    "generation_bytes_per_second": 458714.96968259953,
+    "upload_bytes_per_second": 5273739.379642944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.39345657922652,
+    "estimated_full_shards_per_hour": 98.42955415590752,
     "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": [
@@ -214,6 +214,12 @@
       "bytes": 10758587,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 0.9081728179999118,
+      "bytes": 2358450,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -22,3 +22,4 @@
 {"step": 527200, "loss": 0.5773270130157471, "lr": 0.0003, "grad_norm": 0.167763305804286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77164356970, "tokens_per_second": 150880.05545506466, "data_wait_s": 8.96299980013282e-06, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:8f1d9442fd6ca9ed7634f22d8dd1477afe7d49e94ae78a6a728fa0e725c1322c
-size 50997
+oid sha256:f5ff6265cb45151eb2ad9666e92a5795ce11720ec4a591958eafe8e228e77c62
+size 53115