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": 77295.42005337025,
-      "required_pages_per_second_upper_estimate": 0.0053474765093762975
+      "required_token_positions_per_second": 77121.55787925642,
+      "required_pages_per_second_upper_estimate": 0.005335448320755279
     },
     {
       "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": 9661.927506671282,
-      "required_pages_per_second_upper_estimate": 0.004045247283053578
+      "required_token_positions_per_second": 9640.194734907052,
+      "required_pages_per_second_upper_estimate": 0.004036148225347777
     },
     {
       "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": 19323.855013342563,
-      "required_pages_per_second_upper_estimate": 0.004024649394953147
+      "required_token_positions_per_second": 19280.389469814105,
+      "required_pages_per_second_upper_estimate": 0.004015596668499636
     },
     {
       "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": 28985.782520013847,
-      "required_pages_per_second_upper_estimate": 0.0017376534033475854
+      "required_token_positions_per_second": 28920.58420472116,
+      "required_pages_per_second_upper_estimate": 0.001733744863897852
     },
     {
       "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": 28985.782520013847,
-      "required_pages_per_second_upper_estimate": 0.0022535170959478565
+      "required_token_positions_per_second": 28920.58420472116,
+      "required_pages_per_second_upper_estimate": 0.002248448213710931
     },
     {
       "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": 28985.782520013847,
-      "required_pages_per_second_upper_estimate": 0.37179849565826306
+      "required_token_positions_per_second": 28920.58420472116,
+      "required_pages_per_second_upper_estimate": 0.3709622016741853
     }
   ],
   "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": 0.1569016787734654,
-    "producer_p95_seconds": 0.35645790000126,
-    "gpu_update_seconds": 0.9538469413723729,
-    "producer_batches_per_second": 6.3734181037272375,
-    "gpu_batches_per_second": 1.0483862311926304,
-    "producer_headroom": 6.079265364327535,
+    "producer_batch_seconds": 0.10591865535116085,
+    "producer_p95_seconds": 0.3685349919978762,
+    "gpu_update_seconds": 0.9559972856802313,
+    "producer_batches_per_second": 9.441207468926201,
+    "gpu_batches_per_second": 1.0460280745341854,
+    "producer_headroom": 9.025768713837376,
     "sustainable": true,
-    "buffer_seconds": 60.09235730645949,
+    "buffer_seconds": 60.22782899785457,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330482.60199680825,
-    "upload_bytes_per_second": 4518421.695680749,
+    "generation_bytes_per_second": 333474.96293069696,
+    "upload_bytes_per_second": 4422029.498676617,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.9138731472796,
+    "estimated_full_shards_per_hour": 71.55596414509469,
     "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": 1.4380262719932944,
-      "bytes": 12216092,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.1287659260051441,
       "bytes": 715029,
@@ -562,6 +556,12 @@
       "bytes": 11578062,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.0980768339941278,
+      "bytes": 1310271,
+      "pages": 37,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -427,3 +427,4 @@
 {"step": 647100, "loss": 0.5414162874221802, "lr": 0.0001, "grad_norm": 0.18974881924892584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94718643414, "tokens_per_second": 151372.93778320594, "data_wait_s": 1.3470002159010619e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 647200, "loss": 0.5620044469833374, "lr": 0.0001, "grad_norm": 0.18821107202987994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94733241049, "tokens_per_second": 151293.08493227622, "data_wait_s": 1.3423996279016137e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 647300, "loss": 0.554100513458252, "lr": 0.0001, "grad_norm": 0.19221046863183036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94747831707, "tokens_per_second": 151829.38303588063, "data_wait_s": 1.6685997252352536e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 647400, "loss": 0.4948877990245819, "lr": 0.0001, "grad_norm": 0.17518709669107033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94762596298, "tokens_per_second": 152200.82729375848, "data_wait_s": 1.3095996109768748e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc3f95377dd4d69a1f0bf593787390e34b05c8e42ad6b836188089656b1f34f2
-size 915834
+oid sha256:6fcf44ccf0235998d0bb524d70e4c3a9d1289ca5fd8d91dca357858298f07980
+size 917952