Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76062.84522783225,
-      "required_pages_per_second_upper_estimate": 0.006592934644829746
+      "required_token_positions_per_second": 76094.78849141047,
+      "required_pages_per_second_upper_estimate": 0.006595703405957244
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9507.855653479031,
-      "required_pages_per_second_upper_estimate": 0.0037928257752828432
+      "required_token_positions_per_second": 9511.848561426308,
+      "required_pages_per_second_upper_estimate": 0.003794418605962306
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19015.711306958063,
-      "required_pages_per_second_upper_estimate": 0.004398950145775978
+      "required_token_positions_per_second": 19023.697122852616,
+      "required_pages_per_second_upper_estimate": 0.004400797523737623
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28523.566960437092,
-      "required_pages_per_second_upper_estimate": 0.0017369355763004618
+      "required_token_positions_per_second": 28535.545684278928,
+      "required_pages_per_second_upper_estimate": 0.0017376650177349227
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28523.566960437092,
-      "required_pages_per_second_upper_estimate": 0.002195823142901349
+      "required_token_positions_per_second": 28535.545684278928,
+      "required_pages_per_second_upper_estimate": 0.0021967452982219237
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28523.566960437092,
-      "required_pages_per_second_upper_estimate": 0.3541672394109179
+      "required_token_positions_per_second": 28535.545684278928,
+      "required_pages_per_second_upper_estimate": 0.3543159750708237
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.21397182537504023,
-    "producer_p95_seconds": 0.21227809399897524,
-    "gpu_update_seconds": 0.9693037353409717,
-    "producer_batches_per_second": 4.673512497485334,
-    "gpu_batches_per_second": 1.0316683651778462,
-    "producer_headroom": 4.530053120975248,
+    "producer_batch_seconds": 0.0850086379609678,
+    "producer_p95_seconds": 0.20910069099954853,
+    "gpu_update_seconds": 0.9688968385571157,
+    "producer_batches_per_second": 11.76351043830576,
+    "gpu_batches_per_second": 1.0321016234186533,
+    "producer_headroom": 11.39762807400808,
     "sustainable": true,
-    "buffer_seconds": 60.096831591140244,
+    "buffer_seconds": 60.07160399054118,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 462672.86511479365,
-    "upload_bytes_per_second": 5309734.899886867,
+    "generation_bytes_per_second": 484271.7804400362,
+    "upload_bytes_per_second": 5318222.561147836,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.27882638056619,
+    "estimated_full_shards_per_hour": 103.91345081234755,
     "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": 1.2336509919996388,
-      "bytes": 6127983,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0918112409999594,
       "bytes": 968274,
@@ -562,6 +556,12 @@
       "bytes": 11311428,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.1716579149997415,
+      "bytes": 6474393,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -128,3 +128,4 @@
 {"step": 167800, "loss": 0.6374548450112343, "lr": 0.0003, "grad_norm": 0.16923725250118113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24558282554, "tokens_per_second": 149795.56277722394, "data_wait_s": 1.1805999747593887e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 167900, "loss": 0.5849163699895144, "lr": 0.0003, "grad_norm": 0.1398667099829425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24572900308, "tokens_per_second": 149857.7722695745, "data_wait_s": 1.554099981149193e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 168000, "loss": 0.5603897906839848, "lr": 0.0003, "grad_norm": 0.13752290221883634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24587342803, "tokens_per_second": 147960.3790001725, "data_wait_s": 1.2880000213044696e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 168100, "loss": 0.5387906767427921, "lr": 0.0003, "grad_norm": 0.1564837904446995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24602123606, "tokens_per_second": 150961.6990159981, "data_wait_s": 1.2391001291689463e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c3214d5f13798f9b2f2ad64fbd440526299a267d4d41a414b744a835774ccf37
-size 277071
+oid sha256:add2d9499c5487ede0c1278c15a76ed924099ada4d2103efda642188824a98f5
+size 279189