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": 74463.18105675664,
-      "required_pages_per_second_upper_estimate": 0.006454279808792687
+      "required_token_positions_per_second": 75842.65035729513,
+      "required_pages_per_second_upper_estimate": 0.006573848711530385
     },
     {
       "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": 9307.89763209458,
-      "required_pages_per_second_upper_estimate": 0.003713059530913747
+      "required_token_positions_per_second": 9480.331294661892,
+      "required_pages_per_second_upper_estimate": 0.003781845897024849
     },
     {
       "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": 18615.79526418916,
-      "required_pages_per_second_upper_estimate": 0.0043064366075633576
+      "required_token_positions_per_second": 18960.662589323783,
+      "required_pages_per_second_upper_estimate": 0.004386215593775627
     },
     {
       "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": 27923.69289628374,
-      "required_pages_per_second_upper_estimate": 0.0017004063930895015
+      "required_token_positions_per_second": 28440.993883985673,
+      "required_pages_per_second_upper_estimate": 0.0017319073091720209
     },
     {
       "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": 27923.69289628374,
-      "required_pages_per_second_upper_estimate": 0.0021496431768851343
+      "required_token_positions_per_second": 28440.993883985673,
+      "required_pages_per_second_upper_estimate": 0.0021894664388992136
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27923.69289628374,
-      "required_pages_per_second_upper_estimate": 0.3467188111834777
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28440.993883985673,
+      "required_pages_per_second_upper_estimate": 0.35314195815570076
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06401931735148025,
-    "producer_p95_seconds": 0.20038754100096412,
-    "gpu_update_seconds": 0.9901269184807417,
-    "producer_batches_per_second": 15.62028527279943,
-    "gpu_batches_per_second": 1.0099715312602626,
-    "producer_headroom": 15.466064922947012,
+    "producer_batch_seconds": 0.11872671528128365,
+    "producer_p95_seconds": 0.24425809999956982,
+    "gpu_update_seconds": 0.9721179264261863,
+    "producer_batches_per_second": 8.422704170925904,
+    "gpu_batches_per_second": 1.028681781104806,
+    "producer_headroom": 8.18786171354168,
     "sustainable": true,
-    "buffer_seconds": 60.39774202732524,
+    "buffer_seconds": 60.27131143842355,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 507571.50512871635,
-    "upload_bytes_per_second": 4696580.57876751,
+    "generation_bytes_per_second": 486245.73672719684,
+    "upload_bytes_per_second": 4706554.383009884,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.913029340707,
+    "estimated_full_shards_per_hour": 104.33701588022164,
     "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": 1.096408940999936,
-      "bytes": 6091544,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.042788374000338,
       "bytes": 6163555,
@@ -562,6 +556,12 @@
       "bytes": 830160,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.2181768209993606,
+      "bytes": 7539545,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -144,3 +144,4 @@
 {"step": 251500, "loss": 0.5889512877911329, "lr": 0.0003, "grad_norm": 0.14478115157437446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36812610155, "tokens_per_second": 149492.76537624534, "data_wait_s": 1.1166999684064649e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 251600, "loss": 0.5790707748383284, "lr": 0.0003, "grad_norm": 0.15698482256301324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36827321479, "tokens_per_second": 150597.5952132396, "data_wait_s": 1.3327999113244005e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 251700, "loss": 0.5901068933308125, "lr": 0.0003, "grad_norm": 0.14446392576909967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36841982178, "tokens_per_second": 149373.30216558228, "data_wait_s": 1.1337999239913188e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 251800, "loss": 0.5683508180081844, "lr": 0.0003, "grad_norm": 0.13879556538568094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36856778963, "tokens_per_second": 151435.23461506402, "data_wait_s": 9.588000466465019e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04f29e0ffcdf9542caeced6a13801fa552548077db737d26514a9013c9c47072
-size 311742
+oid sha256:2df2658a2d2322325d73c4ea480979781b5c9ee658a7cc2d4f311e4a43fd399b
+size 313860