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": 76233.76985198172,
-      "required_pages_per_second_upper_estimate": 0.0052740290850763856
+      "required_token_positions_per_second": 76503.02555541412,
+      "required_pages_per_second_upper_estimate": 0.0052926568193991455
     },
     {
       "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": 9529.221231497715,
-      "required_pages_per_second_upper_estimate": 0.003989685936860558
+      "required_token_positions_per_second": 9562.878194426765,
+      "required_pages_per_second_upper_estimate": 0.004003777404401642
     },
     {
       "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": 19058.44246299543,
-      "required_pages_per_second_upper_estimate": 0.003969370960115482
+      "required_token_positions_per_second": 19125.75638885353,
+      "required_pages_per_second_upper_estimate": 0.003983390675684122
     },
     {
       "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": 28587.66369449315,
-      "required_pages_per_second_upper_estimate": 0.0017137867876498682
+      "required_token_positions_per_second": 28688.634583280294,
+      "required_pages_per_second_upper_estimate": 0.0017198398382590372
     },
     {
       "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": 28587.66369449315,
-      "required_pages_per_second_upper_estimate": 0.002222565108403269
+      "required_token_positions_per_second": 28688.634583280294,
+      "required_pages_per_second_upper_estimate": 0.0022304151508824666
     },
     {
       "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": 28587.66369449315,
-      "required_pages_per_second_upper_estimate": 0.3666918548311739
+      "required_token_positions_per_second": 28688.634583280294,
+      "required_pages_per_second_upper_estimate": 0.3679870009784417
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.2877758333121392,
-    "producer_p95_seconds": 0.33934007700008806,
-    "gpu_update_seconds": 0.9671304481354258,
-    "producer_batches_per_second": 3.474926954395573,
-    "gpu_batches_per_second": 1.0339866787649432,
-    "producer_headroom": 3.3607076626424615,
+    "producer_batch_seconds": 0.10618602864019522,
+    "producer_p95_seconds": 0.3297105389938224,
+    "gpu_update_seconds": 0.9637265907424268,
+    "producer_batches_per_second": 9.417434786910038,
+    "gpu_batches_per_second": 1.0376386929716541,
+    "producer_headroom": 9.075832320727944,
     "sustainable": true,
-    "buffer_seconds": 60.929218232531824,
+    "buffer_seconds": 60.71477521677289,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 534758.941983755,
-    "upload_bytes_per_second": 3812859.1356116687,
+    "generation_bytes_per_second": 533908.2626709718,
+    "upload_bytes_per_second": 3823224.3195770737,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.74682039865958,
+    "estimated_full_shards_per_hour": 114.56428442093721,
     "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": 1.6795359120005742,
-      "bytes": 5921876,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.7577662710027653,
       "bytes": 6866333,
@@ -562,6 +556,12 @@
       "bytes": 6221682,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.4849864080024417,
+      "bytes": 6331011,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -359,3 +359,4 @@
 {"step": 835900, "loss": 0.5728225111961365, "lr": 3e-05, "grad_norm": 0.305666108864444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122352146307, "tokens_per_second": 152935.41326874128, "data_wait_s": 7.442002242896706e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 836000, "loss": 0.5252567529678345, "lr": 3e-05, "grad_norm": 0.2783521958592825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122366882287, "tokens_per_second": 152712.3218189268, "data_wait_s": 8.353999874088913e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 836100, "loss": 0.521915078163147, "lr": 3e-05, "grad_norm": 0.2953872750197862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122381410408, "tokens_per_second": 151258.0742203701, "data_wait_s": 1.55930028995499e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 836200, "loss": 0.5268641114234924, "lr": 3e-05, "grad_norm": 0.29490348842268344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122395894812, "tokens_per_second": 149424.52945566384, "data_wait_s": 1.2730000889860094e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bba96d9513185a20454430009edfbe0db7f3a4ef92a6e0bfd84b65bdef090d7d
-size 770244
+oid sha256:15c62e5052e8e98c775b8b0a0ed6ece64eb9bb68f138b1bc0b9b03fb90329e2c
+size 772362