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": 77421.48013735052,
-      "required_pages_per_second_upper_estimate": 0.005356197638485745
+      "required_token_positions_per_second": 77382.36746132474,
+      "required_pages_per_second_upper_estimate": 0.00535349173280437
     },
     {
       "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": 9677.685017168815,
-      "required_pages_per_second_upper_estimate": 0.004051844623644658
+      "required_token_positions_per_second": 9672.795932665593,
+      "required_pages_per_second_upper_estimate": 0.004049797666058853
     },
     {
       "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": 19355.37003433763,
-      "required_pages_per_second_upper_estimate": 0.004031213142719426
+      "required_token_positions_per_second": 19345.591865331186,
+      "required_pages_per_second_upper_estimate": 0.004029176607983042
     },
     {
       "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": 29033.055051506446,
-      "required_pages_per_second_upper_estimate": 0.0017404873194295906
+      "required_token_positions_per_second": 29018.38779799678,
+      "required_pages_per_second_upper_estimate": 0.0017396080399772855
     },
     {
       "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": 29033.055051506446,
-      "required_pages_per_second_upper_estimate": 0.0022571923273415145
+      "required_token_positions_per_second": 29018.38779799678,
+      "required_pages_per_second_upper_estimate": 0.002256052012895568
     },
     {
       "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": 29033.055051506446,
-      "required_pages_per_second_upper_estimate": 0.3724048569349604
+      "required_token_positions_per_second": 29018.38779799678,
+      "required_pages_per_second_upper_estimate": 0.3722167211554082
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.16130055517174924,
-    "producer_p95_seconds": 0.3426047999964794,
-    "gpu_update_seconds": 0.9522938578441272,
-    "producer_batches_per_second": 6.199606684150729,
-    "gpu_batches_per_second": 1.050096030508769,
-    "producer_headroom": 5.9038473663661355,
+    "producer_batch_seconds": 0.0918971643985742,
+    "producer_p95_seconds": 0.3413023940011044,
+    "gpu_update_seconds": 0.9527751918012695,
+    "producer_batches_per_second": 10.881728577204228,
+    "gpu_batches_per_second": 1.0495655308881937,
+    "producer_headroom": 10.367841032275114,
     "sustainable": true,
-    "buffer_seconds": 60.94680690202414,
+    "buffer_seconds": 60.02483708347997,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347668.6676327744,
-    "upload_bytes_per_second": 5825998.701305492,
+    "generation_bytes_per_second": 358200.07912809716,
+    "upload_bytes_per_second": 5758438.43568379,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.60160276162551,
+    "estimated_full_shards_per_hour": 76.86139850980936,
     "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": 1.25489443799961,
-      "bytes": 9999755,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 0.9152432709997811,
       "bytes": 861201,
@@ -562,6 +556,12 @@
       "bytes": 6125208,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 0.8623744039941812,
+      "bytes": 2649650,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -345,3 +345,4 @@
 {"step": 519500, "loss": 0.6142432689666748, "lr": 0.0003, "grad_norm": 0.17161350796879724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76037130595, "tokens_per_second": 151493.0708625333, "data_wait_s": 1.553299807710573e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 519600, "loss": 0.5893015265464783, "lr": 0.0003, "grad_norm": 0.16470768422338858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76051784898, "tokens_per_second": 152697.7118825172, "data_wait_s": 1.1382006050553173e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 519700, "loss": 0.6022124886512756, "lr": 0.0003, "grad_norm": 0.15502248068019933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76066605419, "tokens_per_second": 154577.72313474957, "data_wait_s": 1.563599653309211e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 519800, "loss": 0.6011606454849243, "lr": 0.0003, "grad_norm": 0.16392495698426476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76081282985, "tokens_per_second": 153256.86754452638, "data_wait_s": 1.2156997399870306e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eff990c1e0574f8874dbe1803974267eb4916dfc5f7d0e7880660941285b32ee
-size 739809
+oid sha256:9ef5e369c776968cc8aae704510608d4f53f2cf098ec681467c0d0165c4c17db
+size 741927