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": 77125.94494110333,
-      "required_pages_per_second_upper_estimate": 0.005335751827873233
+      "required_token_positions_per_second": 76824.19326070412,
+      "required_pages_per_second_upper_estimate": 0.005314875946462855
     },
     {
       "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": 9640.743117637916,
-      "required_pages_per_second_upper_estimate": 0.004036377821745656
+      "required_token_positions_per_second": 9603.024157588015,
+      "required_pages_per_second_upper_estimate": 0.004020585680834216
     },
     {
       "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": 19281.486235275832,
-      "required_pages_per_second_upper_estimate": 0.004015825095821663
+      "required_token_positions_per_second": 19206.04831517603,
+      "required_pages_per_second_upper_estimate": 0.00400011336649661
     },
     {
       "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": 28922.229352913746,
-      "required_pages_per_second_upper_estimate": 0.001733843488020002
+      "required_token_positions_per_second": 28809.072472764045,
+      "required_pages_per_second_upper_estimate": 0.0017270599058355788
     },
     {
       "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": 28922.229352913746,
-      "required_pages_per_second_upper_estimate": 0.0022485761167466626
+      "required_token_positions_per_second": 28809.072472764045,
+      "required_pages_per_second_upper_estimate": 0.002239778667039538
     },
     {
       "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": 28922.229352913746,
-      "required_pages_per_second_upper_estimate": 0.37098330386877726
+      "required_token_positions_per_second": 28809.072472764045,
+      "required_pages_per_second_upper_estimate": 0.36953184890860874
     }
   ],
   "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": 0.07226056461719566,
-    "producer_p95_seconds": 0.18103725699984352,
-    "gpu_update_seconds": 0.9559429068428511,
-    "producer_batches_per_second": 13.838806896923037,
-    "gpu_batches_per_second": 1.0460875778686975,
-    "producer_headroom": 13.229109292281503,
+    "producer_batch_seconds": 0.1041843618437781,
+    "producer_p95_seconds": 0.2069937710002705,
+    "gpu_update_seconds": 0.9596976794771259,
+    "producer_batches_per_second": 9.59836948945827,
+    "gpu_batches_per_second": 1.0419948087660607,
+    "producer_headroom": 9.211532925797146,
     "sustainable": true,
-    "buffer_seconds": 60.22440313109962,
+    "buffer_seconds": 60.460953807058935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318520.3588321485,
-    "upload_bytes_per_second": 5740767.590585454,
+    "generation_bytes_per_second": 354670.85882463853,
+    "upload_bytes_per_second": 5772375.171741699,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.3470542309126,
+    "estimated_full_shards_per_hour": 76.10410998873107,
     "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": 1.0081327930001862,
-      "bytes": 1129039,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.1534891419998985,
       "bytes": 7764027,
@@ -562,6 +556,12 @@
       "bytes": 12120623,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.8672846369991021,
+      "bytes": 2691618,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -88,3 +88,4 @@
 {"step": 613200, "loss": 0.5482064485549927, "lr": 0.0001, "grad_norm": 0.14933810984877444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89754711284, "tokens_per_second": 153687.61228747087, "data_wait_s": 9.2249993031146e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 613300, "loss": 0.5628738403320312, "lr": 0.0001, "grad_norm": 0.16078197875598066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89769230678, "tokens_per_second": 150523.0842167897, "data_wait_s": 1.2274000255274586e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 613400, "loss": 0.5531539916992188, "lr": 0.0001, "grad_norm": 0.1555066020701014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89783593944, "tokens_per_second": 149440.4933042486, "data_wait_s": 1.984899972740095e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 613500, "loss": 0.545706033706665, "lr": 0.0001, "grad_norm": 0.1709391725872942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89798140374, "tokens_per_second": 150981.13755542465, "data_wait_s": 9.43500162975397e-06, "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 sha256:9af04bd7c986776f6a849b82e302547f254982f4425c60fbd810a8299d64a178
-size 192351
+oid sha256:fbdf3cbda9e871084bd3e2240702c1c6eeec359e14b297ef31548d1b10b1f62e
+size 194469