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": 76558.28754662313,
-      "required_pages_per_second_upper_estimate": 0.005296479972176471
+      "required_token_positions_per_second": 76255.50143964749,
+      "required_pages_per_second_upper_estimate": 0.005275532526735334
     },
     {
       "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": 9569.785943327892,
-      "required_pages_per_second_upper_estimate": 0.004006669534616344
+      "required_token_positions_per_second": 9531.937679955936,
+      "required_pages_per_second_upper_estimate": 0.00399082325710412
     },
     {
       "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": 19139.571886655784,
-      "required_pages_per_second_upper_estimate": 0.003986268079537154
+      "required_token_positions_per_second": 19063.87535991187,
+      "required_pages_per_second_upper_estimate": 0.00397050248926805
     },
     {
       "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": 28709.357829983677,
-      "required_pages_per_second_upper_estimate": 0.0017210821652563395
+      "required_token_positions_per_second": 28595.813039867808,
+      "required_pages_per_second_upper_estimate": 0.001714275328461749
     },
     {
       "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": 28709.357829983677,
-      "required_pages_per_second_upper_estimate": 0.0022320262921617256
+      "required_token_positions_per_second": 28595.813039867808,
+      "required_pages_per_second_upper_estimate": 0.002223198684160967
     },
     {
       "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": 28709.357829983677,
-      "required_pages_per_second_upper_estimate": 0.368252816536264
+      "required_token_positions_per_second": 28595.813039867808,
+      "required_pages_per_second_upper_estimate": 0.3667963858835547
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07489641312498918,
-    "producer_p95_seconds": 0.18983890299932682,
-    "gpu_update_seconds": 0.9630309449528959,
-    "producer_batches_per_second": 13.35177424760212,
-    "gpu_batches_per_second": 1.038388231697905,
-    "producer_headroom": 12.85817177046601,
+    "producer_batch_seconds": 0.06287530977336075,
+    "producer_p95_seconds": 0.2078481529988494,
+    "gpu_update_seconds": 0.966854831560607,
+    "producer_batches_per_second": 15.904494206145188,
+    "gpu_batches_per_second": 1.0342814322868854,
+    "producer_headroom": 15.377337066739155,
     "sustainable": true,
-    "buffer_seconds": 60.67094953203244,
+    "buffer_seconds": 60.91185438831824,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302968.0764447923,
+    "generation_bytes_per_second": 325488.75312133785,
     "upload_bytes_per_second": 5482064.587464745,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.00989646919086,
+    "estimated_full_shards_per_hour": 69.84230942945577,
     "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": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -88,3 +88,4 @@
 {"step": 768800, "loss": 0.5474236011505127, "lr": 3e-05, "grad_norm": 0.2541554534120326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112529532882, "tokens_per_second": 152436.10863425204, "data_wait_s": 1.85289991350146e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 768900, "loss": 0.5399148464202881, "lr": 3e-05, "grad_norm": 0.24447142968711158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112544190672, "tokens_per_second": 151908.01766839164, "data_wait_s": 1.2266000339877792e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 769000, "loss": 0.5726956129074097, "lr": 3e-05, "grad_norm": 0.24596094504859978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112558881027, "tokens_per_second": 152140.10119972232, "data_wait_s": 1.1895999705302529e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 769100, "loss": 0.5401884317398071, "lr": 3e-05, "grad_norm": 0.24275489347306103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112573549843, "tokens_per_second": 152211.04967640928, "data_wait_s": 8.38999949337449e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7e48c10926458f44eb08eff26b10bbb5bcaabf191afbc09887128818caacdc47
-size 191568
+oid sha256:de2f5d6e2026f3914243d32d108c8be1502907085edb1ed714454ee26a78d04f
+size 193686