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": 77125.41820844865,
-      "required_pages_per_second_upper_estimate": 0.005335715387285995
+      "required_token_positions_per_second": 75949.8795514541,
+      "required_pages_per_second_upper_estimate": 0.005254388895369634
     },
     {
       "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": 9640.677276056082,
-      "required_pages_per_second_upper_estimate": 0.00403635025525027
+      "required_token_positions_per_second": 9493.734943931762,
+      "required_pages_per_second_upper_estimate": 0.003974828569294644
     },
     {
       "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": 19281.354552112163,
-      "required_pages_per_second_upper_estimate": 0.004015797669691389
+      "required_token_positions_per_second": 18987.469887863524,
+      "required_pages_per_second_upper_estimate": 0.003954589244388176
     },
     {
       "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": 28922.031828168245,
-      "required_pages_per_second_upper_estimate": 0.0017338316467131118
+      "required_token_positions_per_second": 28481.204831795287,
+      "required_pages_per_second_upper_estimate": 0.0017074047413844025
     },
     {
       "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": 28922.031828168245,
-      "required_pages_per_second_upper_estimate": 0.0022485607600665194
+      "required_token_positions_per_second": 28481.204831795287,
+      "required_pages_per_second_upper_estimate": 0.00221428840009156
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28922.031828168245,
-      "required_pages_per_second_upper_estimate": 0.37098077023342757
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28481.204831795287,
+      "required_pages_per_second_upper_estimate": 0.3653263148471067
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.22782729545310332,
-    "producer_p95_seconds": 0.18474830699960876,
-    "gpu_update_seconds": 0.9559494355120853,
-    "producer_batches_per_second": 4.389289694244924,
-    "gpu_batches_per_second": 1.0460804335998353,
-    "producer_headroom": 4.195939005512449,
+    "producer_batch_seconds": 0.3509143740077292,
+    "producer_p95_seconds": 0.1661253089996535,
+    "gpu_update_seconds": 0.9707454499654758,
+    "producer_batches_per_second": 2.8496980291208422,
+    "gpu_batches_per_second": 1.0301361701314846,
+    "producer_headroom": 2.766331395544642,
     "sustainable": true,
-    "buffer_seconds": 60.22481443726138,
+    "buffer_seconds": 60.1862178978595,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438039.28333854047,
-    "upload_bytes_per_second": 4331069.976195898,
+    "generation_bytes_per_second": 395974.3986401875,
+    "upload_bytes_per_second": 4433267.7616018625,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.9930331718174,
+    "estimated_full_shards_per_hour": 84.96688813595027,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 897206,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.459628982999675,
+      "bytes": 13449017,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 848100, "loss": 0.5064298510551453, "lr": 3e-05, "grad_norm": 0.2911187229154403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124138304065, "tokens_per_second": 152643.5207272398, "data_wait_s": 1.2854000488005113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 848200, "loss": 0.5381395816802979, "lr": 3e-05, "grad_norm": 0.28488509242539006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124152970850, "tokens_per_second": 152565.69994112532, "data_wait_s": 1.4092000128584914e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 848300, "loss": 0.5027939081192017, "lr": 3e-05, "grad_norm": 0.2984285320279911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124167642263, "tokens_per_second": 152824.67051482873, "data_wait_s": 7.983000614331104e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 848400, "loss": 0.5915848016738892, "lr": 3e-05, "grad_norm": 0.2935924442510267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124182260231, "tokens_per_second": 152495.19183081103, "data_wait_s": 7.327000275836326e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6efab3e62975efdd2939aded1a94ad98a60c809f22c0e16aa2f70663256b2b36
-size 123792
+oid sha256:263c8b20df4e12d64d91693619b04836e659041aaaede1c54b931c1c65bd06e5
+size 125910