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": 76902.77786500512,
-      "required_pages_per_second_upper_estimate": 0.005320312611729801
+      "required_token_positions_per_second": 76902.77470230925,
+      "required_pages_per_second_upper_estimate": 0.005320312392927163
     },
     {
       "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": 9612.84723312564,
-      "required_pages_per_second_upper_estimate": 0.004024698397432676
+      "required_token_positions_per_second": 9612.846837788657,
+      "required_pages_per_second_upper_estimate": 0.004024698231913337
     },
     {
       "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": 19225.69446625128,
-      "required_pages_per_second_upper_estimate": 0.00400420514166174
+      "required_token_positions_per_second": 19225.693675577313,
+      "required_pages_per_second_upper_estimate": 0.0040042049769852045
     },
     {
       "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": 28838.54169937692,
-      "required_pages_per_second_upper_estimate": 0.0017288265409741172
+      "required_token_positions_per_second": 28838.54051336597,
+      "required_pages_per_second_upper_estimate": 0.0017288264698745717
     },
     {
       "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": 28838.54169937692,
-      "required_pages_per_second_upper_estimate": 0.00224206976978726
+      "required_token_positions_per_second": 28838.54051336597,
+      "required_pages_per_second_upper_estimate": 0.0022420696775801254
     },
     {
       "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": 28838.54169937692,
-      "required_pages_per_second_upper_estimate": 0.3699098485060084
+      "required_token_positions_per_second": 28838.54051336597,
+      "required_pages_per_second_upper_estimate": 0.36990983329313337
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.33708331570309724,
-    "producer_p95_seconds": 0.16761950600084674,
-    "gpu_update_seconds": 0.9587169936750775,
-    "producer_batches_per_second": 2.9666256186965936,
-    "gpu_batches_per_second": 1.0430606806776954,
-    "producer_headroom": 2.8441543945162646,
+    "producer_batch_seconds": 0.052329313710934855,
+    "producer_p95_seconds": 0.15855196899974544,
+    "gpu_update_seconds": 0.9587170331031772,
+    "producer_batches_per_second": 19.109748037667035,
+    "gpu_batches_per_second": 1.0430606377808873,
+    "producer_headroom": 18.320840942021402,
     "sustainable": true,
-    "buffer_seconds": 60.39917060152988,
+    "buffer_seconds": 60.399173085500166,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416771.8243569367,
-    "upload_bytes_per_second": 3422087.484360947,
+    "generation_bytes_per_second": 416455.525025813,
+    "upload_bytes_per_second": 3476176.657096584,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.42953155547214,
+    "estimated_full_shards_per_hour": 89.3616610820846,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 7230449,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0110029770003166,
+      "bytes": 6117097,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -46,3 +46,4 @@
 {"step": 529600, "loss": 0.5900040864944458, "lr": 0.0003, "grad_norm": 0.17206534298399126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77516048745, "tokens_per_second": 151456.50016157047, "data_wait_s": 1.1076000191678759e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 529700, "loss": 0.5776903629302979, "lr": 0.0003, "grad_norm": 0.16426839846052832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77530597653, "tokens_per_second": 151204.83870092846, "data_wait_s": 1.902499934658408e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 529800, "loss": 0.6108845472335815, "lr": 0.0003, "grad_norm": 0.16569644422253524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77545158333, "tokens_per_second": 151146.28189862586, "data_wait_s": 1.0017000022344291e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 529900, "loss": 0.552044153213501, "lr": 0.0003, "grad_norm": 0.16098867098718225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77559841799, "tokens_per_second": 153285.25726875255, "data_wait_s": 1.6678999600117095e-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 sha256:232618b3bf4bfc173c7eb19f9baf8ea818a661414ffb08237732b7b7466782c1
-size 101829
+oid sha256:2ea0d2cbc64a84a3909308062c3856b59862f80c990342f315a8d169d8b8b788
+size 103947