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": 77633.80783329153,
-      "required_pages_per_second_upper_estimate": 0.005370886961159076
+      "required_token_positions_per_second": 77669.76179468862,
+      "required_pages_per_second_upper_estimate": 0.005373374339633211
     },
     {
       "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": 9704.225979161442,
-      "required_pages_per_second_upper_estimate": 0.00406295677019266
+      "required_token_positions_per_second": 9708.720224336077,
+      "required_pages_per_second_upper_estimate": 0.004064838416796768
     },
     {
       "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": 19408.451958322883,
-      "required_pages_per_second_upper_estimate": 0.004042268707621082
+      "required_token_positions_per_second": 19417.440448672154,
+      "required_pages_per_second_upper_estimate": 0.004044140773118405
     },
     {
       "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": 29112.677937484328,
-      "required_pages_per_second_upper_estimate": 0.0017452605898668574
+      "required_token_positions_per_second": 29126.16067300823,
+      "required_pages_per_second_upper_estimate": 0.0017460688592745703
     },
     {
       "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": 29112.677937484328,
-      "required_pages_per_second_upper_estimate": 0.0022633826565022323
+      "required_token_positions_per_second": 29126.16067300823,
+      "required_pages_per_second_upper_estimate": 0.0022644308798849297
     },
     {
       "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": 29112.677937484328,
-      "required_pages_per_second_upper_estimate": 0.3734261738238905
+      "required_token_positions_per_second": 29126.16067300823,
+      "required_pages_per_second_upper_estimate": 0.37359911587855765
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.21441338510956598,
-    "producer_p95_seconds": 0.20712211699901673,
-    "gpu_update_seconds": 0.9496893435695084,
-    "producer_batches_per_second": 4.663887935396368,
-    "gpu_batches_per_second": 1.052975909197205,
-    "producer_headroom": 4.429244671848326,
+    "producer_batch_seconds": 0.05773259109373896,
+    "producer_p95_seconds": 0.1975079759995424,
+    "gpu_update_seconds": 0.9492497246855446,
+    "producer_batches_per_second": 17.321238854087202,
+    "gpu_batches_per_second": 1.0534635660086888,
+    "producer_headroom": 16.442181213454834,
     "sustainable": true,
-    "buffer_seconds": 60.78011798844854,
+    "buffer_seconds": 60.75198237987485,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 510774.27871433523,
-    "upload_bytes_per_second": 5372392.324397347,
+    "generation_bytes_per_second": 526097.592953524,
+    "upload_bytes_per_second": 5413361.26720654,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.60026999542754,
+    "estimated_full_shards_per_hour": 112.88829652265825,
     "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": 1.1274267459998555,
-      "bytes": 2324297,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.315684739000062,
       "bytes": 5833727,
@@ -562,6 +556,12 @@
       "bytes": 2581068,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.9553985610000382,
+      "bytes": 10094743,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -114,3 +114,4 @@
 {"step": 896400, "loss": 0.52915358543396, "lr": 3e-05, "grad_norm": 0.31634236585155595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131207877377, "tokens_per_second": 152045.3253269531, "data_wait_s": 1.336299828835763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 896500, "loss": 0.5308711528778076, "lr": 3e-05, "grad_norm": 0.31997333905983366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131222365870, "tokens_per_second": 151356.98767846482, "data_wait_s": 1.1264999557170086e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 896600, "loss": 0.4638233482837677, "lr": 3e-05, "grad_norm": 0.31239623884435447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131236821018, "tokens_per_second": 151455.73450354926, "data_wait_s": 1.4345001545734704e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 896700, "loss": 0.49390313029289246, "lr": 3e-05, "grad_norm": 0.3204091288738879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131251416954, "tokens_per_second": 151661.19877466385, "data_wait_s": 1.4450999515247531e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e852269e2d0a0d246ae7877b16b922f31a1edd69de8d4c6d2a962f0812736e0
-size 247419
+oid sha256:746ca33e536b8dfc86923492b62f5528a83406707a702976ff224cb177b259d8
+size 249537