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": 76549.55702930156,
-      "required_pages_per_second_upper_estimate": 0.005295875974730573
+      "required_token_positions_per_second": 76698.42225402225,
+      "required_pages_per_second_upper_estimate": 0.005306174816391661
     },
     {
       "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": 9568.694628662695,
-      "required_pages_per_second_upper_estimate": 0.004006212623955268
+      "required_token_positions_per_second": 9587.302781752782,
+      "required_pages_per_second_upper_estimate": 0.004014003469071664
     },
     {
       "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": 19137.38925732539,
-      "required_pages_per_second_upper_estimate": 0.003985813495407439
+      "required_token_positions_per_second": 19174.605563505564,
+      "required_pages_per_second_upper_estimate": 0.003993564670524775
     },
     {
       "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": 28706.083885988086,
-      "required_pages_per_second_upper_estimate": 0.0017208858973128268
+      "required_token_positions_per_second": 28761.908345258347,
+      "required_pages_per_second_upper_estimate": 0.0017242324884070652
     },
     {
       "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": 28706.083885988086,
-      "required_pages_per_second_upper_estimate": 0.002231771757416626
+      "required_token_positions_per_second": 28761.908345258347,
+      "required_pages_per_second_upper_estimate": 0.0022361118635790427
     },
     {
       "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": 28706.083885988086,
-      "required_pages_per_second_upper_estimate": 0.3682108218979757
+      "required_token_positions_per_second": 28761.908345258347,
+      "required_pages_per_second_upper_estimate": 0.36892687812186026
     }
   ],
   "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": 0.11230419489848487,
-    "producer_p95_seconds": 0.20592548999957216,
-    "gpu_update_seconds": 0.9631407791396948,
-    "producer_batches_per_second": 8.904386883356672,
-    "gpu_batches_per_second": 1.038269816478157,
-    "producer_headroom": 8.576178120597424,
+    "producer_batch_seconds": 0.10285139768754448,
+    "producer_p95_seconds": 0.18996654600050533,
+    "gpu_update_seconds": 0.961271403417083,
+    "producer_batches_per_second": 9.722765295207088,
+    "gpu_batches_per_second": 1.0402889303117169,
+    "producer_headroom": 9.346216240418627,
     "sustainable": true,
-    "buffer_seconds": 60.67786908580077,
+    "buffer_seconds": 60.56009841527622,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 243630.47636648206,
-    "upload_bytes_per_second": 5021021.942177789,
+    "generation_bytes_per_second": 260639.45202113842,
+    "upload_bytes_per_second": 5026052.852406001,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.277428801020115,
+    "estimated_full_shards_per_hour": 55.92715902782073,
     "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": 0.9162838960000954,
-      "bytes": 5018838,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.177252060999308,
       "bytes": 8046612,
@@ -562,6 +556,12 @@
       "bytes": 6251371,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1527972080002655,
+      "bytes": 6634416,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -120,3 +120,4 @@
 {"step": 452200, "loss": 0.6027845144271851, "lr": 0.0003, "grad_norm": 0.2114486650853434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66188068098, "tokens_per_second": 151629.75374200332, "data_wait_s": 1.3932000001659617e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 452300, "loss": 0.5852214694023132, "lr": 0.0003, "grad_norm": 0.14925860282673115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66202700341, "tokens_per_second": 152224.62511914797, "data_wait_s": 1.639299989619758e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 452400, "loss": 0.5918483734130859, "lr": 0.0003, "grad_norm": 0.16275686269325879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66217395782, "tokens_per_second": 153387.65632560238, "data_wait_s": 1.3433000276563689e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 452500, "loss": 0.639524519443512, "lr": 0.0003, "grad_norm": 0.16123465768285003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66232162314, "tokens_per_second": 153681.12839323672, "data_wait_s": 1.3325001418706961e-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 sha256:18e6ce5732728d5e90cc609814dd2228c39e01e0972e4810ae6bfe7a3ca4a8ab
-size 260127
+oid sha256:162c3b9908e26659eda5486fa6145b1f1c220a05edf15e5d868aaa9f606b261e
+size 262245