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": 77258.50733977498,
-      "required_pages_per_second_upper_estimate": 0.005344922802200481
+      "required_token_positions_per_second": 77585.75495287833,
+      "required_pages_per_second_upper_estimate": 0.005367562551394046
     },
     {
       "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": 9657.313417471873,
-      "required_pages_per_second_upper_estimate": 0.004043315460258925
+      "required_token_positions_per_second": 9698.219369109791,
+      "required_pages_per_second_upper_estimate": 0.004060441927996315
     },
     {
       "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": 19314.626834943745,
-      "required_pages_per_second_upper_estimate": 0.004022727408756104
+      "required_token_positions_per_second": 19396.438738219582,
+      "required_pages_per_second_upper_estimate": 0.004039766670683484
     },
     {
       "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": 28971.940252415618,
-      "required_pages_per_second_upper_estimate": 0.0017368235805409905
+      "required_token_positions_per_second": 29094.658107329375,
+      "required_pages_per_second_upper_estimate": 0.0017441803285637553
     },
     {
       "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": 28971.940252415618,
-      "required_pages_per_second_upper_estimate": 0.0022524409205277947
+      "required_token_positions_per_second": 29094.658107329375,
+      "required_pages_per_second_upper_estimate": 0.0022619816939685408
     },
     {
       "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": 28971.940252415618,
-      "required_pages_per_second_upper_estimate": 0.371620941912182
+      "required_token_positions_per_second": 29094.658107329375,
+      "required_pages_per_second_upper_estimate": 0.3731950347908489
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05356916924998245,
-    "producer_p95_seconds": 0.18347788500068418,
-    "gpu_update_seconds": 0.9543026721413582,
-    "producer_batches_per_second": 18.66745394787558,
-    "gpu_batches_per_second": 1.0478855704722083,
-    "producer_headroom": 17.814401184533413,
+    "producer_batch_seconds": 0.07160311682029885,
+    "producer_p95_seconds": 0.18674470800033305,
+    "gpu_update_seconds": 0.950277535415859,
+    "producer_batches_per_second": 13.965872498395333,
+    "gpu_batches_per_second": 1.0523241502940313,
+    "producer_headroom": 13.271454897707242,
     "sustainable": true,
-    "buffer_seconds": 60.12106834490557,
+    "buffer_seconds": 60.81776226661498,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378317.2010819194,
-    "upload_bytes_per_second": 5830039.400269368,
+    "generation_bytes_per_second": 395539.98417073773,
+    "upload_bytes_per_second": 5847066.041321276,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.1780645784682,
+    "estimated_full_shards_per_hour": 84.87367290345763,
     "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": 1.2162191410006926,
-      "bytes": 6401722,
-      "pages": 45,
-      "shards": 2
-    },
     {
       "seconds": 1.3737986060004914,
       "bytes": 7819084,
@@ -562,6 +556,12 @@
       "bytes": 10485207,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.0128666689997772,
+      "bytes": 6455974,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -106,3 +106,4 @@
 {"step": 535600, "loss": 0.6079011559486389, "lr": 0.0003, "grad_norm": 0.15751586357119715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78394365462, "tokens_per_second": 151797.8055164574, "data_wait_s": 1.3263999790069647e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 535700, "loss": 0.582923948764801, "lr": 0.0003, "grad_norm": 0.1584891579298035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78408957915, "tokens_per_second": 152132.8347766694, "data_wait_s": 1.3298000340000726e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 535800, "loss": 0.5835523009300232, "lr": 0.0003, "grad_norm": 0.18165182247936754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78423673544, "tokens_per_second": 153891.21582447164, "data_wait_s": 1.8556000213720836e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 535900, "loss": 0.5594803690910339, "lr": 0.0003, "grad_norm": 0.1546807488656789, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78438554401, "tokens_per_second": 155480.88742217954, "data_wait_s": 1.7360000128974207e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1556adc69afa601367e685f8194a2972cb4e9f02d38a9b680b4d3440e474f774
-size 230475
+oid sha256:bdb2687c7cd7078a7f13dfae99c3a5e1f7e7c2a5475c717a4471a9cefc8a1e7c
+size 232593