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": 76503.2879975074,
-      "required_pages_per_second_upper_estimate": 0.005292674975752109
+      "required_token_positions_per_second": 75740.9409478639,
+      "required_pages_per_second_upper_estimate": 0.005239934037968911
     },
     {
       "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": 9562.910999688425,
-      "required_pages_per_second_upper_estimate": 0.004003791139279648
+      "required_token_positions_per_second": 9467.617618482987,
+      "required_pages_per_second_upper_estimate": 0.003963893790521015
     },
     {
       "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": 19125.82199937685,
-      "required_pages_per_second_upper_estimate": 0.003983404340625865
+      "required_token_positions_per_second": 18935.235236965975,
+      "required_pages_per_second_upper_estimate": 0.003943710144126595
     },
     {
       "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": 28688.73299906528,
-      "required_pages_per_second_upper_estimate": 0.001719845738135075
+      "required_token_positions_per_second": 28402.852855448964,
+      "required_pages_per_second_upper_estimate": 0.001702707660038981
     },
     {
       "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": 28688.73299906528,
-      "required_pages_per_second_upper_estimate": 0.002230422802276864
+      "required_token_positions_per_second": 28402.852855448964,
+      "required_pages_per_second_upper_estimate": 0.0022081968785645553
     },
     {
       "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": 28688.73299906528,
-      "required_pages_per_second_upper_estimate": 0.36798826335046086
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28402.852855448964,
+      "required_pages_per_second_upper_estimate": 0.3643212998223338
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10190685189849091,
-    "producer_p95_seconds": 0.23042260199872544,
-    "gpu_update_seconds": 0.9637232847090463,
-    "producer_batches_per_second": 9.812882856945643,
-    "gpu_batches_per_second": 1.0376422525703588,
-    "producer_headroom": 9.456903699360746,
+    "producer_batch_seconds": 0.09125689333603759,
+    "producer_p95_seconds": 0.24653962600132218,
+    "gpu_update_seconds": 0.97342334379963,
+    "producer_batches_per_second": 10.958076299153364,
+    "gpu_batches_per_second": 1.027302258949977,
+    "producer_headroom": 10.66684727273334,
     "sustainable": true,
-    "buffer_seconds": 60.71456693666992,
+    "buffer_seconds": 60.35224731557706,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283882.00208504975,
-    "upload_bytes_per_second": 5035383.836630575,
+    "generation_bytes_per_second": 296201.5031420495,
+    "upload_bytes_per_second": 5003765.548838004,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.91446921266193,
+    "estimated_full_shards_per_hour": 63.55794735618699,
     "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": 1.3445512510006665,
-      "bytes": 5828365,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.2113412039998366,
       "bytes": 6839007,
@@ -562,6 +556,12 @@
       "bytes": 6221386,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2352482690002944,
+      "bytes": 2570616,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -179,3 +179,4 @@
 {"step": 737900, "loss": 0.5347310304641724, "lr": 5e-05, "grad_norm": 0.2168170256007732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108008061585, "tokens_per_second": 150672.47344936585, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738000, "loss": 0.4385727047920227, "lr": 5e-05, "grad_norm": 0.22462039073004075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108022860653, "tokens_per_second": 152859.93455801453, "data_wait_s": 1.8253998860018328e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738100, "loss": 0.535503625869751, "lr": 5e-05, "grad_norm": 0.2167565098519966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108037535724, "tokens_per_second": 150431.75698752122, "data_wait_s": 1.7815997125580907e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 738200, "loss": 0.5587160587310791, "lr": 5e-05, "grad_norm": 0.2236781793592614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108052271803, "tokens_per_second": 132744.06884565984, "data_wait_s": 1.8234000890515745e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ad8bb5799d85a7d037350418ba20af12e46598bfea06d4a60559a010cca57a0
-size 385872
+oid sha256:cb1326ad8d7c5473e637b50b49bfa37341c23c84b67ac3e0a62aff27ef832d64
+size 387990