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": 76797.58282488566,
-      "required_pages_per_second_upper_estimate": 0.005313034974768215
+      "required_token_positions_per_second": 77045.93035983328,
+      "required_pages_per_second_upper_estimate": 0.005330216233481564
     },
     {
       "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": 9599.697853110707,
-      "required_pages_per_second_upper_estimate": 0.004019193026610703
+      "required_token_positions_per_second": 9630.74129497916,
+      "required_pages_per_second_upper_estimate": 0.004032190267460242
     },
     {
       "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": 19199.395706221414,
-      "required_pages_per_second_upper_estimate": 0.00399872780349246
+      "required_token_positions_per_second": 19261.48258995832,
+      "required_pages_per_second_upper_estimate": 0.004011658864033627
     },
     {
       "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": 28799.093559332123,
-      "required_pages_per_second_upper_estimate": 0.0017264616852122538
+      "required_token_positions_per_second": 28892.223884937477,
+      "required_pages_per_second_upper_estimate": 0.0017320447060305212
     },
     {
       "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": 28799.093559332123,
-      "required_pages_per_second_upper_estimate": 0.002239002850412808
+      "required_token_positions_per_second": 28892.223884937477,
+      "required_pages_per_second_upper_estimate": 0.0022462433235916133
     },
     {
       "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": 28799.093559332123,
-      "required_pages_per_second_upper_estimate": 0.36940385012162646
+      "required_token_positions_per_second": 28892.223884937477,
+      "required_pages_per_second_upper_estimate": 0.3705984259429391
     }
   ],
   "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": 0.25807861913278884,
-    "producer_p95_seconds": 0.14699011099946802,
-    "gpu_update_seconds": 0.9600302156399253,
-    "producer_batches_per_second": 3.874788246156383,
-    "gpu_batches_per_second": 1.041633881630936,
-    "producer_headroom": 3.71991379551656,
+    "producer_batch_seconds": 0.05059119808597501,
+    "producer_p95_seconds": 0.14752641700033564,
+    "gpu_update_seconds": 0.9569356831134714,
+    "producer_batches_per_second": 19.76628421213891,
+    "gpu_batches_per_second": 1.0450023106531205,
+    "producer_headroom": 18.915062685158173,
     "sustainable": true,
-    "buffer_seconds": 60.481903585315294,
+    "buffer_seconds": 60.2869480361487,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 71,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 327336.0598810324,
-    "upload_bytes_per_second": 5564567.110805273,
+    "generation_bytes_per_second": 327638.5352300076,
+    "upload_bytes_per_second": 5941093.0690124575,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.23869845698574,
+    "estimated_full_shards_per_hour": 70.30360262561008,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 6858425,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.2630921189993387,
+      "bytes": 11470878,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -14,3 +14,4 @@
 {"step": 486400, "loss": 0.6144828796386719, "lr": 0.0003, "grad_norm": 0.15597788152546366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71192908927, "tokens_per_second": 153212.1731888354, "data_wait_s": 9.916999260894954e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 486500, "loss": 0.6309879422187805, "lr": 0.0003, "grad_norm": 0.15481581152410082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71207610359, "tokens_per_second": 153431.4123156743, "data_wait_s": 1.1533000360941514e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 486600, "loss": 0.5777984857559204, "lr": 0.0003, "grad_norm": 0.16132067041184758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71222190790, "tokens_per_second": 152174.30810914058, "data_wait_s": 1.1974999324593227e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 486700, "loss": 0.596780002117157, "lr": 0.0003, "grad_norm": 0.16100248803824965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71236686033, "tokens_per_second": 150745.9795879053, "data_wait_s": 1.3244999536254909e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d39049395bce529eb000440bad1795a33d32119e5c90dd2c27db240c751bfd80
-size 34053
+oid sha256:c6902ffbbaad6e952949f2043c3b83e8414dadc4d8ac92ea0491745b81b80921
+size 36171