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": 77157.61092278478,
-      "required_pages_per_second_upper_estimate": 0.0053379425539611565
+      "required_token_positions_per_second": 76368.90562931377,
+      "required_pages_per_second_upper_estimate": 0.00528337809171042
     },
     {
       "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": 9644.701365348097,
-      "required_pages_per_second_upper_estimate": 0.004038035057404297
+      "required_token_positions_per_second": 9546.113203664221,
+      "required_pages_per_second_upper_estimate": 0.003996758252862183
     },
     {
       "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": 19289.402730696194,
-      "required_pages_per_second_upper_estimate": 0.0040174738930457005
+      "required_token_positions_per_second": 19092.226407328442,
+      "required_pages_per_second_upper_estimate": 0.003976407264777522
     },
     {
       "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": 28934.10409604429,
-      "required_pages_per_second_upper_estimate": 0.0017345553607389951
+      "required_token_positions_per_second": 28638.33961099266,
+      "required_pages_per_second_upper_estimate": 0.0017168247314663697
     },
     {
       "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": 28934.10409604429,
-      "required_pages_per_second_upper_estimate": 0.0022494993257935877
+      "required_token_positions_per_second": 28638.33961099266,
+      "required_pages_per_second_upper_estimate": 0.0022265049380113065
     },
     {
       "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": 28934.10409604429,
-      "required_pages_per_second_upper_estimate": 0.37113562032355013
+      "required_token_positions_per_second": 28638.33961099266,
+      "required_pages_per_second_upper_estimate": 0.3673418710764698
     }
   ],
   "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": 0.12439553868765074,
-    "producer_p95_seconds": 0.2759809920025873,
-    "gpu_update_seconds": 0.9555505817019535,
-    "producer_batches_per_second": 8.038873504225391,
-    "gpu_batches_per_second": 1.0465170752330835,
-    "producer_headroom": 7.681550253190993,
+    "producer_batch_seconds": 0.09293254514074079,
+    "producer_p95_seconds": 0.2693924419982068,
+    "gpu_update_seconds": 0.9654190981584517,
+    "producer_batches_per_second": 10.760492984299093,
+    "gpu_batches_per_second": 1.0358195750503711,
+    "producer_headroom": 10.388385432642377,
     "sustainable": true,
-    "buffer_seconds": 60.19968664722307,
+    "buffer_seconds": 60.82140318398246,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 425525.201446124,
-    "upload_bytes_per_second": 3234298.338676098,
+    "generation_bytes_per_second": 409211.03159911674,
+    "upload_bytes_per_second": 3199947.680312593,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.30780251062193,
+    "estimated_full_shards_per_hour": 87.8071614358914,
     "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": 2.000728038001398,
-      "bytes": 11892571,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.4507886779974797,
       "bytes": 2119215,
@@ -562,6 +556,12 @@
       "bytes": 7589463,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.5132595840004797,
+      "bytes": 5701329,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -275,3 +275,4 @@
 {"step": 827500, "loss": 0.48631832003593445, "lr": 3e-05, "grad_norm": 0.2867443741708145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121122404899, "tokens_per_second": 151536.75105540224, "data_wait_s": 9.069000952877104e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 827600, "loss": 0.5027718544006348, "lr": 3e-05, "grad_norm": 0.28944081476856853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121137128326, "tokens_per_second": 152636.3020098213, "data_wait_s": 1.4097000530455261e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 827700, "loss": 0.5382410883903503, "lr": 3e-05, "grad_norm": 0.2850816636258624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121151734190, "tokens_per_second": 149867.66891448153, "data_wait_s": 9.880997822619975e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 827800, "loss": 0.5772397518157959, "lr": 3e-05, "grad_norm": 0.30076087097337906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121166486746, "tokens_per_second": 152557.78921766672, "data_wait_s": 1.3916000170866027e-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 sha256:0f62e244048c6b6cd4dc1a0d0eb79fda247d6c2af719eea2de466368a38239c1
-size 590766
+oid sha256:c2ed1fe27782f909633fdcd6a82bbbb57317bd648eaf3d2c085edec0e219f5cb
+size 592884