Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75974.16159674301,
-      "required_pages_per_second_upper_estimate": 0.0065852477724535384
+      "required_token_positions_per_second": 75906.42456455444,
+      "required_pages_per_second_upper_estimate": 0.006579376498181372
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9496.770199592876,
-      "required_pages_per_second_upper_estimate": 0.0037884036219853505
+      "required_token_positions_per_second": 9488.303070569305,
+      "required_pages_per_second_upper_estimate": 0.0037850259576229874
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18993.540399185753,
-      "required_pages_per_second_upper_estimate": 0.0043938212964575035
+      "required_token_positions_per_second": 18976.60614113861,
+      "required_pages_per_second_upper_estimate": 0.00438990385389106
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28490.310598778626,
-      "required_pages_per_second_upper_estimate": 0.001734910438358106
+      "required_token_positions_per_second": 28464.909211707913,
+      "required_pages_per_second_upper_estimate": 0.001733363627156278
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28490.310598778626,
-      "required_pages_per_second_upper_estimate": 0.002193262976121379
+      "required_token_positions_per_second": 28464.909211707913,
+      "required_pages_per_second_upper_estimate": 0.0021913075070291357
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28490.310598778626,
-      "required_pages_per_second_upper_estimate": 0.35375430670100233
+      "required_token_positions_per_second": 28464.909211707913,
+      "required_pages_per_second_upper_estimate": 0.3534389064865579
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.2736097927422918,
-    "producer_p95_seconds": 0.36162391000107164,
-    "gpu_update_seconds": 0.970435190734118,
-    "producer_batches_per_second": 3.6548399455200866,
-    "gpu_batches_per_second": 1.0304655164488796,
-    "producer_headroom": 3.546785299633459,
+    "producer_batch_seconds": 0.1345641699296607,
+    "producer_p95_seconds": 0.34754705000159447,
+    "gpu_update_seconds": 0.9713011833049546,
+    "producer_batches_per_second": 7.4313987187133055,
+    "gpu_batches_per_second": 1.0295467741503153,
+    "producer_headroom": 7.218126369097157,
     "sustainable": true,
-    "buffer_seconds": 60.166981825515315,
+    "buffer_seconds": 60.220673364907185,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380513.39012413076,
-    "upload_bytes_per_second": 5580629.210244831,
+    "generation_bytes_per_second": 423350.7582627978,
+    "upload_bytes_per_second": 5543080.375712826,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.6493156222624,
+    "estimated_full_shards_per_hour": 90.84121762192679,
     "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": 1.248066406998987,
-      "bytes": 7262597,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.8794477230003395,
       "bytes": 663465,
@@ -562,6 +556,12 @@
       "bytes": 11665620,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.8683807099987462,
+      "bytes": 2266346,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -266,3 +266,4 @@
 {"step": 181600, "loss": 0.5649946480989456, "lr": 0.0003, "grad_norm": 0.13432296823074194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26578620979, "tokens_per_second": 149636.97172450728, "data_wait_s": 7.533999450970441e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 181700, "loss": 0.6719467882066965, "lr": 0.0003, "grad_norm": 0.15690616034954094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26593333780, "tokens_per_second": 149614.2150639295, "data_wait_s": 1.3469998521031812e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 181800, "loss": 0.5850318782031536, "lr": 0.0003, "grad_norm": 0.14054630884511277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26607924724, "tokens_per_second": 148969.4430190043, "data_wait_s": 1.3633998605655506e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 181900, "loss": 0.5988446213304996, "lr": 0.0003, "grad_norm": 0.13855545873119954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26622431128, "tokens_per_second": 147935.4875856118, "data_wait_s": 1.4346001989906654e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d63ee2555f4db9b8c4ce7a6b671d182e236bcff0dd08b54a573c3e548ffecd1
-size 571704
+oid sha256:dd40d7ab9db9f52bc0b615ee097e6fbd155e77697e2ffb6f0e640cdc394ece47
+size 573822