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": 77358.71220521323,
-      "required_pages_per_second_upper_estimate": 0.0053518552072988205
+      "required_token_positions_per_second": 77218.30391105988,
+      "required_pages_per_second_upper_estimate": 0.005342141435717163
     },
     {
       "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": 9669.839025651654,
-      "required_pages_per_second_upper_estimate": 0.004048559670839357
+      "required_token_positions_per_second": 9652.287988882485,
+      "required_pages_per_second_upper_estimate": 0.00404121141825143
     },
     {
       "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": 19339.678051303308,
-      "required_pages_per_second_upper_estimate": 0.0040279449164787985
+      "required_token_positions_per_second": 19304.57597776497,
+      "required_pages_per_second_upper_estimate": 0.004020634080264693
     },
     {
       "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": 29009.517076954962,
-      "required_pages_per_second_upper_estimate": 0.0017390762537956353
+      "required_token_positions_per_second": 28956.863966647452,
+      "required_pages_per_second_upper_estimate": 0.0017359197802293443
     },
     {
       "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": 29009.517076954962,
-      "required_pages_per_second_upper_estimate": 0.002255362353352745
+      "required_token_positions_per_second": 28956.863966647452,
+      "required_pages_per_second_upper_estimate": 0.0022512688056228925
     },
     {
       "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": 29009.517076954962,
-      "required_pages_per_second_upper_estimate": 0.37210293707052194
+      "required_token_positions_per_second": 28956.863966647452,
+      "required_pages_per_second_upper_estimate": 0.37142755950600237
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1503987471481878,
-    "producer_p95_seconds": 0.30198867499711923,
-    "gpu_update_seconds": 0.9530665376695793,
-    "producer_batches_per_second": 6.648991557188309,
-    "gpu_batches_per_second": 1.0492446859431048,
-    "producer_headroom": 6.3369313624037265,
+    "producer_batch_seconds": 0.086731079195431,
+    "producer_p95_seconds": 0.30754697900556494,
+    "gpu_update_seconds": 0.9547995263521974,
+    "producer_batches_per_second": 11.529892274794618,
+    "gpu_batches_per_second": 1.0473402765714501,
+    "producer_headroom": 11.008735682865762,
     "sustainable": true,
-    "buffer_seconds": 60.043191873183495,
+    "buffer_seconds": 60.15237016018844,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 41751.62114795341,
-    "upload_bytes_per_second": 5432193.636886988,
+    "generation_bytes_per_second": 41718.82809784483,
+    "upload_bytes_per_second": 5443661.699790895,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 8.95892597035362,
+    "estimated_full_shards_per_hour": 8.951889345183455,
     "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": [
-    {
-      "seconds": 1.2183306759980042,
-      "bytes": 7063267,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.0798259360017255,
       "bytes": 4403386,
@@ -562,6 +556,12 @@
       "bytes": 10646786,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1512752330017975,
+      "bytes": 7594320,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -341,3 +341,4 @@
 {"step": 719100, "loss": 0.5993387699127197, "lr": 8e-05, "grad_norm": 0.22667573341498234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105256470148, "tokens_per_second": 150256.08197712986, "data_wait_s": 1.6924997908063233e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 719200, "loss": 0.5463426113128662, "lr": 8e-05, "grad_norm": 0.23426403055278208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105271034141, "tokens_per_second": 151377.8714554051, "data_wait_s": 1.4236000424716622e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 719300, "loss": 0.5330504179000854, "lr": 8e-05, "grad_norm": 0.22170238373314438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105285534605, "tokens_per_second": 151215.92307629634, "data_wait_s": 1.2185999366920441e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 719400, "loss": 0.502803385257721, "lr": 8e-05, "grad_norm": 0.20656039022238373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105300180788, "tokens_per_second": 152180.42777615122, "data_wait_s": 1.4493001799564809e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6a74be476157dbc305764b9f5d46e57ad251d1f8539e9b56e974c6ec597704f4
-size 731337
+oid sha256:bc61d1dcda583ed6453481c1bd5c88789a097b8bc552c81d36e85112504004af
+size 733455