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": 76317.81649150292,
-      "required_pages_per_second_upper_estimate": 0.005279843626613534
+      "required_token_positions_per_second": 76098.65846053648,
+      "required_pages_per_second_upper_estimate": 0.005264681765514583
     },
     {
       "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": 9539.727061437865,
-      "required_pages_per_second_upper_estimate": 0.003994084508469822
+      "required_token_positions_per_second": 9512.33230756706,
+      "required_pages_per_second_upper_estimate": 0.003982614897091628
     },
     {
       "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": 19079.45412287573,
-      "required_pages_per_second_upper_estimate": 0.003973747134753815
+      "required_token_positions_per_second": 19024.66461513412,
+      "required_pages_per_second_upper_estimate": 0.003962335925187721
     },
     {
       "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": 28619.181184313595,
-      "required_pages_per_second_upper_estimate": 0.0017156762130401833
+      "required_token_positions_per_second": 28536.99692270118,
+      "required_pages_per_second_upper_estimate": 0.0017107493920446206
     },
     {
       "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": 28619.181184313595,
-      "required_pages_per_second_upper_estimate": 0.002225015454606022
+      "required_token_positions_per_second": 28536.99692270118,
+      "required_pages_per_second_upper_estimate": 0.0022186259897560203
     },
     {
       "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": 28619.181184313595,
-      "required_pages_per_second_upper_estimate": 0.3670961273497466
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28536.99692270118,
+      "required_pages_per_second_upper_estimate": 0.36604195588436755
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08822450099188472,
-    "producer_p95_seconds": 0.31486475199926645,
-    "gpu_update_seconds": 0.9660653748945862,
-    "producer_batches_per_second": 11.334719819973643,
-    "gpu_batches_per_second": 1.0351266342706016,
-    "producer_headroom": 10.950080352207934,
+    "producer_batch_seconds": 0.11071569065580888,
+    "producer_p95_seconds": 0.3281187750035315,
+    "gpu_update_seconds": 0.9688475656668,
+    "producer_batches_per_second": 9.032143448472752,
+    "gpu_batches_per_second": 1.032154113234273,
+    "producer_headroom": 8.750770192806161,
     "sustainable": true,
-    "buffer_seconds": 60.862118618358934,
+    "buffer_seconds": 60.0685490713416,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373179.6714843427,
-    "upload_bytes_per_second": 4933758.640842618,
+    "generation_bytes_per_second": 384852.16732313135,
+    "upload_bytes_per_second": 4899088.933693607,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.07567032239639,
+    "estimated_full_shards_per_hour": 82.58031620760399,
     "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": 1.036487166002189,
-      "bytes": 2690894,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1432905639994715,
       "bytes": 5745742,
@@ -562,6 +556,12 @@
       "bytes": 13422269,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.324093607996474,
+      "bytes": 1086878,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -336,3 +336,4 @@
 {"step": 590700, "loss": 0.6077341437339783, "lr": 0.0003, "grad_norm": 0.15848083511875027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86460672697, "tokens_per_second": 153083.06157210996, "data_wait_s": 1.0315998224541545e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 590800, "loss": 0.5838284492492676, "lr": 0.0003, "grad_norm": 0.17198594263842423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86475194571, "tokens_per_second": 150726.01126480653, "data_wait_s": 1.0618001397233456e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 590900, "loss": 0.5800275802612305, "lr": 0.0003, "grad_norm": 0.18188725123569327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86489847229, "tokens_per_second": 152318.04262182084, "data_wait_s": 7.337002898566425e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 591000, "loss": 0.5615832805633545, "lr": 0.0003, "grad_norm": 0.19398214386955098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86504490178, "tokens_per_second": 152016.46038496826, "data_wait_s": 1.4429002476390451e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5932eb4459f4dea02b1ef4a2589b583dbef5e9b395ae3c9024b804dfe98b0dc
-size 721530
+oid sha256:ecec109f8976e1178e202149f125494f4da229a3ba9c59e4a551689816dbc2f9
+size 723648