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": 76041.62915785471,
-      "required_pages_per_second_upper_estimate": 0.0052607363460288456
+      "required_token_positions_per_second": 75860.06898877802,
+      "required_pages_per_second_upper_estimate": 0.005248175592254487
     },
     {
       "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": 9505.203644731839,
-      "required_pages_per_second_upper_estimate": 0.00397963027482593
+      "required_token_positions_per_second": 9482.508623597252,
+      "required_pages_per_second_upper_estimate": 0.003970128343402811
     },
     {
       "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": 19010.407289463677,
-      "required_pages_per_second_upper_estimate": 0.00395936650024153
+      "required_token_positions_per_second": 18965.017247194504,
+      "required_pages_per_second_upper_estimate": 0.003949912951452772
     },
     {
       "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": 28515.61093419552,
-      "required_pages_per_second_upper_estimate": 0.00170946733468822
+      "required_token_positions_per_second": 28447.525870791756,
+      "required_pages_per_second_upper_estimate": 0.0017053857390970366
     },
     {
       "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": 28515.61093419552,
-      "required_pages_per_second_upper_estimate": 0.0022169633232167274
+      "required_token_positions_per_second": 28447.525870791756,
+      "required_pages_per_second_upper_estimate": 0.0022116700090116293
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28515.61093419552,
-      "required_pages_per_second_upper_estimate": 0.36576763938630236
+      "required_token_positions_per_second": 28447.525870791756,
+      "required_pages_per_second_upper_estimate": 0.36489431729700433
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1245757904842435,
-    "producer_p95_seconds": 0.36707382899476215,
-    "gpu_update_seconds": 0.9695741768886638,
-    "producer_batches_per_second": 8.02724185905512,
-    "gpu_batches_per_second": 1.0313806038120485,
-    "producer_headroom": 7.783006418179596,
+    "producer_batch_seconds": 0.1203577716251516,
+    "producer_p95_seconds": 0.35561221199895954,
+    "gpu_update_seconds": 0.9718947132898941,
+    "producer_batches_per_second": 8.308561935779695,
+    "gpu_batches_per_second": 1.0289180364146324,
+    "producer_headroom": 8.075047420425934,
     "sustainable": true,
-    "buffer_seconds": 60.113598967097154,
+    "buffer_seconds": 60.25747222397343,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426565.46488532197,
-    "upload_bytes_per_second": 4644527.978541236,
+    "generation_bytes_per_second": 446317.0541759361,
+    "upload_bytes_per_second": 4589330.91137192,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.53101882858033,
+    "estimated_full_shards_per_hour": 95.7692500968796,
     "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": 1.4163199819995498,
-      "bytes": 7360891,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.1906762940016051,
       "bytes": 1163380,
@@ -562,6 +556,12 @@
       "bytes": 13332219,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.2068677400020533,
+      "bytes": 1198185,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -395,3 +395,4 @@
 {"step": 643900, "loss": 0.5616509318351746, "lr": 0.0001, "grad_norm": 0.17307198984621422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94250311988, "tokens_per_second": 150143.02719182655, "data_wait_s": 1.323899778071791e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 644000, "loss": 0.5245442986488342, "lr": 0.0001, "grad_norm": 0.17180469405553145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94264816392, "tokens_per_second": 149160.67662374603, "data_wait_s": 9.87599923973903e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 644100, "loss": 0.5561860799789429, "lr": 0.0001, "grad_norm": 0.17385707982534387, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94279316177, "tokens_per_second": 150438.48340043137, "data_wait_s": 1.4274002751335502e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 644200, "loss": 0.6420584917068481, "lr": 0.0001, "grad_norm": 0.17948412050355866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94294021687, "tokens_per_second": 152359.3150423586, "data_wait_s": 1.723800232866779e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f48533299630f1daf22266a1b06be9816db06d7f96199700b8727c98302ad81b
-size 847275
+oid sha256:0c319fecd426835546728451ad251cd591835d80265ea06eca6276abd06c9dd8
+size 849393