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": 75991.40848115578,
-      "required_pages_per_second_upper_estimate": 0.005257261963086789
+      "required_token_positions_per_second": 77513.10158329534,
+      "required_pages_per_second_upper_estimate": 0.005362536222707252
     },
     {
       "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": 9498.926060144473,
-      "required_pages_per_second_upper_estimate": 0.0039770019812500725
+      "required_token_positions_per_second": 9689.137697911918,
+      "required_pages_per_second_upper_estimate": 0.0040566396219126255
     },
     {
       "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": 18997.852120288946,
-      "required_pages_per_second_upper_estimate": 0.003956751589604511
+      "required_token_positions_per_second": 19378.275395823835,
+      "required_pages_per_second_upper_estimate": 0.004035983725462
     },
     {
       "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": 28496.77818043342,
-      "required_pages_per_second_upper_estimate": 0.001708338339856132
+      "required_token_positions_per_second": 29067.41309373575,
+      "required_pages_per_second_upper_estimate": 0.0017425470315995447
     },
     {
       "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": 28496.77818043342,
-      "required_pages_per_second_upper_estimate": 0.0022154991594482533
+      "required_token_positions_per_second": 29067.41309373575,
+      "required_pages_per_second_upper_estimate": 0.0022598635139997867
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28496.77818043342,
-      "required_pages_per_second_upper_estimate": 0.3655260730420777
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29067.41309373575,
+      "required_pages_per_second_upper_estimate": 0.37284556500988636
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08603192614847899,
-    "producer_p95_seconds": 0.2382367329992121,
-    "gpu_update_seconds": 0.9702149423678986,
-    "producer_batches_per_second": 11.623591900919909,
-    "gpu_batches_per_second": 1.0306994422899818,
-    "producer_headroom": 11.277382546258982,
+    "producer_batch_seconds": 0.06681454621102034,
+    "producer_p95_seconds": 0.23892979200172704,
+    "gpu_update_seconds": 0.951168234711549,
+    "producer_batches_per_second": 14.96680074488122,
+    "gpu_batches_per_second": 1.0513387259019007,
+    "producer_headroom": 14.235945443788166,
     "sustainable": true,
-    "buffer_seconds": 60.15332642680971,
+    "buffer_seconds": 60.874767021539135,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291111.3394165927,
-    "upload_bytes_per_second": 5100683.334792493,
+    "generation_bytes_per_second": 291295.9691390401,
+    "upload_bytes_per_second": 5037474.848808913,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.46571671365104,
+    "estimated_full_shards_per_hour": 62.50533395412829,
     "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": 1.1527972080002655,
-      "bytes": 6634416,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.1540112700004101,
       "bytes": 11807839,
@@ -562,6 +556,12 @@
       "bytes": 7390991,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0645652359999076,
+      "bytes": 834051,
+      "pages": 23,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 460100, "loss": 0.549582302570343, "lr": 0.0003, "grad_norm": 0.1420083697173705, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67344454921, "tokens_per_second": 133765.46454767417, "data_wait_s": 1.1686999641824514e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 460200, "loss": 0.5797005295753479, "lr": 0.0003, "grad_norm": 0.1720838286779801, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67359130439, "tokens_per_second": 152246.32421488495, "data_wait_s": 1.3431999832391739e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 460300, "loss": 0.6818230748176575, "lr": 0.0003, "grad_norm": 0.1774673164541843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67373822275, "tokens_per_second": 152216.01214944094, "data_wait_s": 1.3736000255448744e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 460400, "loss": 0.546209990978241, "lr": 0.0003, "grad_norm": 0.15693226266937596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67388442938, "tokens_per_second": 152288.8353781708, "data_wait_s": 1.4305998774943873e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:302bffdd0de53db11915e3f71aa251260974464db98d2717992c2148d681da11
-size 429015
+oid sha256:92e645806b600a02e3aefc2b21a6c1e472e84dfaaeb9944c654a813626918605
+size 431133