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": 77266.82056202172,
-      "required_pages_per_second_upper_estimate": 0.005345497930204843
+      "required_token_positions_per_second": 76824.51523633588,
+      "required_pages_per_second_upper_estimate": 0.005314898221484146
     },
     {
       "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": 9658.352570252715,
-      "required_pages_per_second_upper_estimate": 0.004043750531828286
+      "required_token_positions_per_second": 9603.064404541985,
+      "required_pages_per_second_upper_estimate": 0.004020602531393391
     },
     {
       "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": 19316.70514050543,
-      "required_pages_per_second_upper_estimate": 0.004023160264995999
+      "required_token_positions_per_second": 19206.12880908397,
+      "required_pages_per_second_upper_estimate": 0.004000130131254867
     },
     {
       "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": 28975.05771075814,
-      "required_pages_per_second_upper_estimate": 0.0017370104674085415
+      "required_token_positions_per_second": 28809.193213625953,
+      "required_pages_per_second_upper_estimate": 0.0017270671440658897
     },
     {
       "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": 28975.05771075814,
-      "required_pages_per_second_upper_estimate": 0.0022526832892016757
+      "required_token_positions_per_second": 28809.193213625953,
+      "required_pages_per_second_upper_estimate": 0.0022397880541104686
     },
     {
       "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": 28975.05771075814,
-      "required_pages_per_second_upper_estimate": 0.37166092932053385
+      "required_token_positions_per_second": 28809.193213625953,
+      "required_pages_per_second_upper_estimate": 0.36953339764274384
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11587585167978887,
-    "producer_p95_seconds": 0.2329864989987982,
-    "gpu_update_seconds": 0.9541999976667719,
-    "producer_batches_per_second": 8.6299257826678,
-    "gpu_batches_per_second": 1.0479983257652685,
-    "producer_headroom": 8.234675161686031,
+    "producer_batch_seconds": 0.08509824403139987,
+    "producer_p95_seconds": 0.22491116599849192,
+    "gpu_update_seconds": 0.9596936573330787,
+    "producer_batches_per_second": 11.751123790885934,
+    "gpu_batches_per_second": 1.041999175840059,
+    "producer_headroom": 11.277478968649074,
     "sustainable": true,
-    "buffer_seconds": 60.11459985300663,
+    "buffer_seconds": 60.46070041198396,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309947.9684787567,
-    "upload_bytes_per_second": 5071532.644414549,
+    "generation_bytes_per_second": 295859.7115334075,
+    "upload_bytes_per_second": 4996132.746697853,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.50761881610896,
+    "estimated_full_shards_per_hour": 63.48460683347386,
     "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": 1.2295570830010547,
-      "bytes": 11846540,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 0.9956274190008116,
       "bytes": 6486340,
@@ -562,6 +556,12 @@
       "bytes": 2680179,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.3023044059991662,
+      "bytes": 5788392,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -208,3 +208,4 @@
 {"step": 461000, "loss": 0.5812311172485352, "lr": 0.0003, "grad_norm": 0.15962488188450133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67476569867, "tokens_per_second": 152967.32783971506, "data_wait_s": 1.3201999536249787e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 461100, "loss": 0.5485983490943909, "lr": 0.0003, "grad_norm": 0.1541883087276806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67491107766, "tokens_per_second": 151592.52345898005, "data_wait_s": 1.6062000213423744e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 461200, "loss": 0.5550450682640076, "lr": 0.0003, "grad_norm": 0.16242360347050405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67505691182, "tokens_per_second": 151115.94158695123, "data_wait_s": 1.3955999747850001e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 461300, "loss": 0.541749119758606, "lr": 0.0003, "grad_norm": 0.14559382314573044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67520399023, "tokens_per_second": 152942.82485674313, "data_wait_s": 1.3496999599738047e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2c0e3313a523e3b2e527ae89a82fd6915921915bcc498578e3f2d264a500a9e
-size 448077
+oid sha256:07dcb7666cd4b244eef66c8e4504536b8252d0ef2a317fffb046fb86909587ab
+size 450195