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": 77075.08064539146,
-      "required_pages_per_second_upper_estimate": 0.00533223291787446
+      "required_token_positions_per_second": 76537.66520595078,
+      "required_pages_per_second_upper_estimate": 0.005295053270798334
     },
     {
       "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": 9634.385080673932,
-      "required_pages_per_second_upper_estimate": 0.004033715844439746
+      "required_token_positions_per_second": 9567.208150743847,
+      "required_pages_per_second_upper_estimate": 0.004005590266691835
     },
     {
       "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": 19268.770161347864,
-      "required_pages_per_second_upper_estimate": 0.004013176672967887
+      "required_token_positions_per_second": 19134.416301487694,
+      "required_pages_per_second_upper_estimate": 0.003985194307108565
     },
     {
       "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": 28903.155242021796,
-      "required_pages_per_second_upper_estimate": 0.0017327000241965105
+      "required_token_positions_per_second": 28701.624452231543,
+      "required_pages_per_second_upper_estimate": 0.0017206185610682831
     },
     {
       "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": 28903.155242021796,
-      "required_pages_per_second_upper_estimate": 0.0022470931885229606
+      "required_token_positions_per_second": 28701.624452231543,
+      "required_pages_per_second_upper_estimate": 0.002231425056057031
     },
     {
       "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": 28903.155242021796,
-      "required_pages_per_second_upper_estimate": 0.37073864165444
+      "required_token_positions_per_second": 28701.624452231543,
+      "required_pages_per_second_upper_estimate": 0.36815362106991373
     }
   ],
   "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": 0.059101206296844566,
-    "producer_p95_seconds": 0.2028343820002192,
-    "gpu_update_seconds": 0.9565737639537379,
-    "producer_batches_per_second": 16.920128414593634,
-    "gpu_batches_per_second": 1.0453976867050707,
-    "producer_headroom": 16.185350924128425,
+    "producer_batch_seconds": 0.330018031742199,
+    "producer_p95_seconds": 0.23895061999974132,
+    "gpu_update_seconds": 0.9632904244153462,
+    "producer_batches_per_second": 3.030137458613693,
+    "gpu_batches_per_second": 1.0381085233011986,
+    "producer_headroom": 2.918902398544823,
     "sustainable": true,
-    "buffer_seconds": 60.264147129085494,
+    "buffer_seconds": 60.68729673816681,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380392.40218530095,
-    "upload_bytes_per_second": 5858363.959402369,
+    "generation_bytes_per_second": 356373.9357494258,
+    "upload_bytes_per_second": 5909104.409818855,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.62335442704459,
+    "estimated_full_shards_per_hour": 76.46955065118865,
     "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": 0.95688545899975,
-      "bytes": 1097352,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.40065100099946,
       "bytes": 12237786,
@@ -562,6 +556,12 @@
       "bytes": 6185512,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.246648587000891,
+      "bytes": 6489173,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -130,3 +130,4 @@
 {"step": 498000, "loss": 0.5688769817352295, "lr": 0.0003, "grad_norm": 0.1392845240084225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72890642341, "tokens_per_second": 153795.6250889008, "data_wait_s": 1.0654001016519032e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 498100, "loss": 0.5718763470649719, "lr": 0.0003, "grad_norm": 0.1522275235795989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72905296351, "tokens_per_second": 153234.0952013457, "data_wait_s": 1.0153999028261751e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 498200, "loss": 0.5864498615264893, "lr": 0.0003, "grad_norm": 0.1551933277041455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72919966630, "tokens_per_second": 152959.19352431633, "data_wait_s": 1.6435998986708e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 498300, "loss": 0.5782458782196045, "lr": 0.0003, "grad_norm": 0.1476533573606139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72934606497, "tokens_per_second": 152606.39386160494, "data_wait_s": 9.33699993765913e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:333acb350e38e2bcefcc9d166703dc40f90dec6f6a70821284d6cbfb0429d387
-size 281307
+oid sha256:6ac4a6fc9ee9521f01a173964b2672af597282cc4c330aad80a288938738c786
+size 283425