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": 77139.4963452494,
-      "required_pages_per_second_upper_estimate": 0.005336689345455649
+      "required_token_positions_per_second": 76721.28014490122,
+      "required_pages_per_second_upper_estimate": 0.005307756178320282
     },
     {
       "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": 9642.437043156175,
-      "required_pages_per_second_upper_estimate": 0.004037087032986964
+      "required_token_positions_per_second": 9590.160018112652,
+      "required_pages_per_second_upper_estimate": 0.004015199734269661
     },
     {
       "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": 19284.87408631235,
-      "required_pages_per_second_upper_estimate": 0.00401653069584893
+      "required_token_positions_per_second": 19180.320036225305,
+      "required_pages_per_second_upper_estimate": 0.003994754844491517
     },
     {
       "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": 28927.311129468522,
-      "required_pages_per_second_upper_estimate": 0.001734148132765037
+      "required_token_positions_per_second": 28770.480054337957,
+      "required_pages_per_second_upper_estimate": 0.0017247463492781458
     },
     {
       "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": 28927.311129468522,
-      "required_pages_per_second_upper_estimate": 0.002248971202521424
+      "required_token_positions_per_second": 28770.480054337957,
+      "required_pages_per_second_upper_estimate": 0.0022367782762570186
     },
     {
       "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": 28927.311129468522,
-      "required_pages_per_second_upper_estimate": 0.3710484874420354
+      "required_token_positions_per_second": 28770.480054337957,
+      "required_pages_per_second_upper_estimate": 0.36903682680234934
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.13558026701562653,
-    "producer_p95_seconds": 0.300583035001182,
-    "gpu_update_seconds": 0.955774972525349,
-    "producer_batches_per_second": 7.375704606664798,
-    "gpu_batches_per_second": 1.0462713805508002,
-    "producer_headroom": 7.049513867790138,
+    "producer_batch_seconds": 0.23321185655473187,
+    "producer_p95_seconds": 0.3100268740017782,
+    "gpu_update_seconds": 0.9609850078198917,
+    "producer_batches_per_second": 4.2879466540557845,
+    "gpu_batches_per_second": 1.0405989602986818,
+    "producer_headroom": 4.120652448879076,
     "sustainable": true,
-    "buffer_seconds": 60.21382326909699,
+    "buffer_seconds": 60.542055492653176,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 506736.01526276645,
-    "upload_bytes_per_second": 5739751.965787483,
+    "generation_bytes_per_second": 507987.4198090745,
+    "upload_bytes_per_second": 5719948.047255839,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.73375266468283,
+    "estimated_full_shards_per_hour": 109.00227494911361,
     "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": 1.0163737470029446,
-      "bytes": 7245209,
-      "pages": 27,
-      "shards": 3
-    },
     {
       "seconds": 1.4228277920010441,
       "bytes": 10524529,
@@ -562,6 +556,12 @@
       "bytes": 12857725,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 0.9267861950029328,
+      "bytes": 5234150,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -313,3 +313,4 @@
 {"step": 588400, "loss": 0.5753762125968933, "lr": 0.0003, "grad_norm": 0.17385487216208018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86124218995, "tokens_per_second": 153909.21463850315, "data_wait_s": 1.2128999514970928e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 588500, "loss": 0.5517246127128601, "lr": 0.0003, "grad_norm": 0.16494667235820873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86138952433, "tokens_per_second": 153027.8235048569, "data_wait_s": 8.676001016283408e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 588600, "loss": 0.5555380582809448, "lr": 0.0003, "grad_norm": 0.14995391301965494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86153517225, "tokens_per_second": 151471.8795841019, "data_wait_s": 2.0479001250350848e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 588700, "loss": 0.5562505125999451, "lr": 0.0003, "grad_norm": 0.15879911997357984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86167928257, "tokens_per_second": 149965.56399318564, "data_wait_s": 1.1205000191694126e-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 sha256:12c677da55011ff68675a69715268dff739c2afa55979761bafab9425ea114c0
-size 672033
+oid sha256:72ad7cf4ae574bb7c58b0a494d89646c0021900e33f714ec200e2aab0e90464b
+size 674151