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": 77283.19948499346,
-      "required_pages_per_second_upper_estimate": 0.005346631062100365
+      "required_token_positions_per_second": 77633.72213108835,
+      "required_pages_per_second_upper_estimate": 0.0053708810320818035
     },
     {
       "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": 9660.399935624182,
-      "required_pages_per_second_upper_estimate": 0.0040446077209554685
+      "required_token_positions_per_second": 9704.215266386043,
+      "required_pages_per_second_upper_estimate": 0.004062952284977309
     },
     {
       "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": 19320.799871248364,
-      "required_pages_per_second_upper_estimate": 0.004024013089424441
+      "required_token_positions_per_second": 19408.430532772087,
+      "required_pages_per_second_upper_estimate": 0.004042264245243882
     },
     {
       "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": 28981.199806872548,
-      "required_pages_per_second_upper_estimate": 0.0017373786766921625
+      "required_token_positions_per_second": 29112.64579915813,
+      "required_pages_per_second_upper_estimate": 0.0017452586632232742
     },
     {
       "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": 28981.199806872548,
-      "required_pages_per_second_upper_estimate": 0.002253160810158345
+      "required_token_positions_per_second": 29112.64579915813,
+      "required_pages_per_second_upper_estimate": 0.0022633801578887318
     },
     {
       "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": 28981.199806872548,
-      "required_pages_per_second_upper_estimate": 0.3717397135346205
+      "required_token_positions_per_second": 29112.64579915813,
+      "required_pages_per_second_upper_estimate": 0.3734257615879495
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10717797093764148,
-    "producer_p95_seconds": 0.3102747119992273,
-    "gpu_update_seconds": 0.9539977704250742,
-    "producer_batches_per_second": 9.33027553378317,
-    "gpu_batches_per_second": 1.0482204791258878,
-    "producer_headroom": 8.901062056680765,
+    "producer_batch_seconds": 0.10963028749301884,
+    "producer_p95_seconds": 0.327405488002114,
+    "gpu_update_seconds": 0.9496903919601672,
+    "producer_batches_per_second": 9.121566885097142,
+    "gpu_batches_per_second": 1.05297474678668,
+    "producer_headroom": 8.662664430398786,
     "sustainable": true,
-    "buffer_seconds": 60.10185953677967,
+    "buffer_seconds": 60.780185085450704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290622.52249149367,
-    "upload_bytes_per_second": 5368538.08313504,
+    "generation_bytes_per_second": 289712.85941798077,
+    "upload_bytes_per_second": 5181616.247155882,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.360827980600426,
+    "estimated_full_shards_per_hour": 62.165635460897136,
     "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": 1.3631092460018408,
-      "bytes": 8446837,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.1454860439989716,
       "bytes": 6061904,
@@ -562,6 +556,12 @@
       "bytes": 2733894,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 3.8195672939982614,
+      "bytes": 6121969,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -385,3 +385,4 @@
 {"step": 523500, "loss": 0.6117322444915771, "lr": 0.0003, "grad_norm": 0.15847204102332504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76622393886, "tokens_per_second": 153661.93508374627, "data_wait_s": 1.0378003935329616e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 523600, "loss": 0.5748156905174255, "lr": 0.0003, "grad_norm": 0.15414128055515142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76637125565, "tokens_per_second": 153879.30634632974, "data_wait_s": 1.1898002412635833e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 523700, "loss": 0.5777722597122192, "lr": 0.0003, "grad_norm": 0.15813718426402518, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76651725396, "tokens_per_second": 152906.96139242983, "data_wait_s": 8.773000445216894e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 523800, "loss": 0.5599973201751709, "lr": 0.0003, "grad_norm": 0.16208737839385862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76666386923, "tokens_per_second": 133169.49670360115, "data_wait_s": 1.5987003280315548e-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 sha256:8258cb7e2681cd33f8140e62dc1cd3e82b30d3267db8cee16ad210c471d8019b
-size 825312
+oid sha256:c529830ec6919d79f22038c2e813ece2124899115048b649d395126615e99c99
+size 827430