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": 76824.77270594945,
-      "required_pages_per_second_upper_estimate": 0.005314916033829425
+      "required_token_positions_per_second": 76555.10083167214,
+      "required_pages_per_second_upper_estimate": 0.005296259507841964
     },
     {
       "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": 9603.096588243681,
-      "required_pages_per_second_upper_estimate": 0.00402061600603722
+      "required_token_positions_per_second": 9569.387603959018,
+      "required_pages_per_second_upper_estimate": 0.004006502758240869
     },
     {
       "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": 19206.193176487363,
-      "required_pages_per_second_upper_estimate": 0.004000143537287512
+      "required_token_positions_per_second": 19138.775207918035,
+      "required_pages_per_second_upper_estimate": 0.0039861021523659125
     },
     {
       "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": 28809.289764731046,
-      "required_pages_per_second_upper_estimate": 0.001727072932157215
+      "required_token_positions_per_second": 28708.162811877053,
+      "required_pages_per_second_upper_estimate": 0.0017210105257455344
     },
     {
       "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": 28809.289764731046,
-      "required_pages_per_second_upper_estimate": 0.002239795560534206
+      "required_token_positions_per_second": 28708.162811877053,
+      "required_pages_per_second_upper_estimate": 0.0022319333847602646
     },
     {
       "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": 28809.289764731046,
-      "required_pages_per_second_upper_estimate": 0.3695346360966515
+      "required_token_positions_per_second": 28708.162811877053,
+      "required_pages_per_second_upper_estimate": 0.3682374881271027
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1330073943748289,
-    "producer_p95_seconds": 0.23116804299934302,
-    "gpu_update_seconds": 0.9596904410273689,
-    "producer_batches_per_second": 7.518378994643669,
-    "gpu_batches_per_second": 1.042002667995191,
-    "producer_headroom": 7.215316453180489,
+    "producer_batch_seconds": 0.09270904229683197,
+    "producer_p95_seconds": 0.22362289900047472,
+    "gpu_update_seconds": 0.96307103248563,
+    "producer_batches_per_second": 10.786434367407674,
+    "gpu_batches_per_second": 1.0383450091101365,
+    "producer_headroom": 10.388102483057791,
     "sustainable": true,
-    "buffer_seconds": 60.46049778472424,
+    "buffer_seconds": 60.67347504659469,
     "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": 265871.6938890892,
+    "generation_bytes_per_second": 276561.53500032634,
     "upload_bytes_per_second": 4795241.721250045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.04987633232601,
+    "estimated_full_shards_per_hour": 59.343667388032365,
     "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": [

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -203,3 +203,4 @@
 {"step": 945300, "loss": 0.5615050196647644, "lr": 3e-05, "grad_norm": 0.3714581616074633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138365298444, "tokens_per_second": 154376.5902163234, "data_wait_s": 1.456500103813596e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 945400, "loss": 0.5052629709243774, "lr": 3e-05, "grad_norm": 0.34539745933413624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138379950923, "tokens_per_second": 151624.36703415157, "data_wait_s": 1.2027001503156498e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 945500, "loss": 0.5330240726470947, "lr": 3e-05, "grad_norm": 0.33637224552686323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138394712044, "tokens_per_second": 153378.28099233445, "data_wait_s": 1.1569001799216494e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 945600, "loss": 0.49929627776145935, "lr": 3e-05, "grad_norm": 0.3394370739139647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138409393627, "tokens_per_second": 153207.73386112833, "data_wait_s": 7.845002983231097e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d51f168f19ced394989fab50e290b58840ffcc3c93528737de6700cd0392f0ed
-size 437487
+oid sha256:c30c8b1fcc3351c4f5a6c08c62d94ab5a8feb2d11f34ac3031bae323eb80b391
+size 439605