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": 77232.58938055178,
-      "required_pages_per_second_upper_estimate": 0.005343129737643469
+      "required_token_positions_per_second": 77207.08747691086,
+      "required_pages_per_second_upper_estimate": 0.005341365456777021
     },
     {
       "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": 9654.073672568973,
-      "required_pages_per_second_upper_estimate": 0.004041959046721648
+      "required_token_positions_per_second": 9650.885934613858,
+      "required_pages_per_second_upper_estimate": 0.004040624407407379
     },
     {
       "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": 19308.147345137946,
-      "required_pages_per_second_upper_estimate": 0.0040213779019051925
+      "required_token_positions_per_second": 19301.771869227716,
+      "required_pages_per_second_upper_estimate": 0.004020050058405698
     },
     {
       "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": 28962.221017706917,
-      "required_pages_per_second_upper_estimate": 0.001736240927260616
+      "required_token_positions_per_second": 28952.657803841572,
+      "required_pages_per_second_upper_estimate": 0.0017356676271915706
     },
     {
       "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": 28962.221017706917,
-      "required_pages_per_second_upper_estimate": 0.0022516852927795886
+      "required_token_positions_per_second": 28952.657803841572,
+      "required_pages_per_second_upper_estimate": 0.0022509417949656903
     },
     {
       "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": 28962.221017706917,
-      "required_pages_per_second_upper_estimate": 0.37149627400503993
+      "required_token_positions_per_second": 28952.657803841572,
+      "required_pages_per_second_upper_estimate": 0.3713736073657543
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08003109611698278,
-    "producer_p95_seconds": 0.29185956999936025,
-    "gpu_update_seconds": 0.9546229200825644,
-    "producer_batches_per_second": 12.495143119597955,
-    "gpu_batches_per_second": 1.0475340356520153,
-    "producer_headroom": 11.928150011680163,
+    "producer_batch_seconds": 0.7059129091092871,
+    "producer_p95_seconds": 0.28888258799997857,
+    "gpu_update_seconds": 0.9549382370115788,
+    "producer_batches_per_second": 1.4166053447893292,
+    "gpu_batches_per_second": 1.0471881439468163,
+    "producer_headroom": 1.3527706104943018,
     "sustainable": true,
-    "buffer_seconds": 60.14124396520156,
+    "buffer_seconds": 60.16110893172946,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372323.5771019831,
-    "upload_bytes_per_second": 5837121.971039354,
+    "generation_bytes_per_second": 315298.7111333179,
+    "upload_bytes_per_second": 5815107.049350233,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.89197239679928,
+    "estimated_full_shards_per_hour": 67.65576363086369,
     "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": 1.116415386000881,
-      "bytes": 10937725,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 0.8731399669995881,
       "bytes": 778500,
@@ -562,6 +556,12 @@
       "bytes": 1221938,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.4893900960014435,
+      "bytes": 11508997,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -291,3 +291,4 @@
 {"step": 426200, "loss": 0.6211295127868652, "lr": 0.0003, "grad_norm": 0.17212021409185227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62383928278, "tokens_per_second": 152387.01901556214, "data_wait_s": 8.109996997518465e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 426300, "loss": 0.5704801082611084, "lr": 0.0003, "grad_norm": 0.15625892871854805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62398521600, "tokens_per_second": 152225.03910024115, "data_wait_s": 1.267100014956668e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 426400, "loss": 0.5964219570159912, "lr": 0.0003, "grad_norm": 0.16855262090960713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62413150360, "tokens_per_second": 133586.97409603183, "data_wait_s": 1.0431998816784471e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 426500, "loss": 0.555823028087616, "lr": 0.0003, "grad_norm": 0.1637813321546856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62427777916, "tokens_per_second": 150993.46444015187, "data_wait_s": 1.0121999366674572e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aef551acfa46879b9c953fe8bd42960444b63c551a0bfc35772ed97e9d355a04
-size 625437
+oid sha256:e71ccad7d508bbac1d16151353751ce68a2fc6147a4b6af8d0d3776c54e2adde
+size 627555