Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75844.09042001696,
-      "required_pages_per_second_upper_estimate": 0.0065739735325173075
+      "required_token_positions_per_second": 75934.80441102601,
+      "required_pages_per_second_upper_estimate": 0.006581836391345464
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9480.51130250212,
-      "required_pages_per_second_upper_estimate": 0.003781917704843673
+      "required_token_positions_per_second": 9491.850551378251,
+      "required_pages_per_second_upper_estimate": 0.003786441100757241
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18961.02260500424,
-      "required_pages_per_second_upper_estimate": 0.004386298877067242
+      "required_token_positions_per_second": 18983.701102756502,
+      "required_pages_per_second_upper_estimate": 0.004391545148394299
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28441.53390750636,
-      "required_pages_per_second_upper_estimate": 0.0017319401937711464
+      "required_token_positions_per_second": 28475.551654134753,
+      "required_pages_per_second_upper_estimate": 0.0017340116960634924
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28441.53390750636,
-      "required_pages_per_second_upper_estimate": 0.0021895080114047167
+      "required_token_positions_per_second": 28475.551654134753,
+      "required_pages_per_second_upper_estimate": 0.0021921267917073203
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28441.53390750636,
-      "required_pages_per_second_upper_estimate": 0.35314866344048523
+      "required_token_positions_per_second": 28475.551654134753,
+      "required_pages_per_second_upper_estimate": 0.35357105000353567
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10781340801574402,
-    "producer_p95_seconds": 0.2692386969974905,
-    "gpu_update_seconds": 0.9720994686824213,
-    "producer_batches_per_second": 9.275284200773708,
-    "gpu_batches_per_second": 1.0287013132055252,
-    "producer_headroom": 9.016498843450579,
+    "producer_batch_seconds": 0.11267089006233277,
+    "producer_p95_seconds": 0.28314934599984554,
+    "gpu_update_seconds": 0.9709381695502785,
+    "producer_batches_per_second": 8.875406943592717,
+    "gpu_batches_per_second": 1.0299317004533692,
+    "producer_headroom": 8.617471371825744,
     "sustainable": true,
-    "buffer_seconds": 60.270167058310115,
+    "buffer_seconds": 60.19816651211726,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313041.79635274556,
-    "upload_bytes_per_second": 6411504.493395193,
+    "generation_bytes_per_second": 325517.6887779343,
+    "upload_bytes_per_second": 6441453.476344812,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.17148225724006,
+    "estimated_full_shards_per_hour": 69.84851834777376,
     "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": 0.9357377939995786,
-      "bytes": 1197833,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.0255696429994714,
       "bytes": 13323829,
@@ -562,6 +556,12 @@
       "bytes": 16075425,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.8457192399982887,
+      "bytes": 2628229,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -186,3 +186,4 @@
 {"step": 58600, "loss": 0.6123258080333471, "lr": 0.0003, "grad_norm": 1.2281418722124549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8576757534, "tokens_per_second": 148609.3018100492, "data_wait_s": 9.133000276051462e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 58700, "loss": 0.6020220182836056, "lr": 0.0003, "grad_norm": 0.13453120517679965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8591337147, "tokens_per_second": 149186.54722730984, "data_wait_s": 2.2103999071987346e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 58800, "loss": 0.598078764975071, "lr": 0.0003, "grad_norm": 0.4489301375949946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8605922430, "tokens_per_second": 149153.03213214755, "data_wait_s": 9.407001925865188e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 58900, "loss": 0.618890143930912, "lr": 0.0003, "grad_norm": 0.13939650377971358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8620555221, "tokens_per_second": 149908.78320002125, "data_wait_s": 7.834001735318452e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9fe0958d37aa70117f3cbfe93d58f24928e8a7657afba53bc5788e831041ce7
-size 400698
+oid sha256:f36f45a97d3e8472b76ca1e1b986aed6fa664425c83e55069391e489864d050a
+size 402816