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": 75878.74243770714,
-      "required_pages_per_second_upper_estimate": 0.006576977081585957
+      "required_token_positions_per_second": 75961.49041058178,
+      "required_pages_per_second_upper_estimate": 0.0065841494661519106
     },
     {
       "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": 9484.842804713393,
-      "required_pages_per_second_upper_estimate": 0.003783645605837479
+      "required_token_positions_per_second": 9495.186301322723,
+      "required_pages_per_second_upper_estimate": 0.003787771781283997
     },
     {
       "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": 18969.685609426786,
-      "required_pages_per_second_upper_estimate": 0.004388302910676727
+      "required_token_positions_per_second": 18990.372602645446,
+      "required_pages_per_second_upper_estimate": 0.004393088482479219
     },
     {
       "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": 28454.528414140175,
-      "required_pages_per_second_upper_estimate": 0.0017327314910482383
+      "required_token_positions_per_second": 28485.558903968165,
+      "required_pages_per_second_upper_estimate": 0.001734621085074365
     },
     {
       "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": 28454.528414140175,
-      "required_pages_per_second_upper_estimate": 0.0021905083644964845
+      "required_token_positions_per_second": 28485.558903968165,
+      "required_pages_per_second_upper_estimate": 0.0021928971775013366
     },
     {
       "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": 28454.528414140175,
-      "required_pages_per_second_upper_estimate": 0.35331001172306115
+      "required_token_positions_per_second": 28485.558903968165,
+      "required_pages_per_second_upper_estimate": 0.35369530655435594
     }
   ],
   "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": 0.12244590219518159,
-    "producer_p95_seconds": 0.35823091400015983,
-    "gpu_update_seconds": 0.9716555339662779,
-    "producer_batches_per_second": 8.166871917085285,
-    "gpu_batches_per_second": 1.0291713112753247,
-    "producer_headroom": 7.935386293429702,
+    "producer_batch_seconds": 0.10154058111717745,
+    "producer_p95_seconds": 0.36124689699863666,
+    "gpu_update_seconds": 0.9705970696663603,
+    "producer_batches_per_second": 9.848279269211623,
+    "gpu_batches_per_second": 1.0302936524872746,
+    "producer_headroom": 9.558710999952766,
     "sustainable": true,
-    "buffer_seconds": 60.24264310590923,
+    "buffer_seconds": 60.177018319314335,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 172,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456853.74146765587,
-    "upload_bytes_per_second": 5126569.040437219,
+    "generation_bytes_per_second": 561364.2020831463,
+    "upload_bytes_per_second": 5046862.159451941,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.03017790815599,
+    "estimated_full_shards_per_hour": 60.22784493861576,
     "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": 1.8239342810011294,
-      "bytes": 17807257,
-      "pages": 31,
-      "shards": 5
-    },
     {
       "seconds": 1.1584871390004992,
       "bytes": 5888359,
@@ -562,6 +556,12 @@
       "bytes": 5816616,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.117633730002126,
+      "bytes": 7565346,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -293,3 +293,4 @@
 {"step": 144300, "loss": 0.5910980794578791, "lr": 0.0003, "grad_norm": 0.15748691865434894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21118019654, "tokens_per_second": 149716.23294407126, "data_wait_s": 9.985000360757113e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 144400, "loss": 0.6247450709342957, "lr": 0.0003, "grad_norm": 0.13759672434234607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21132620047, "tokens_per_second": 148990.94409713423, "data_wait_s": 9.557999874232337e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 144500, "loss": 0.6128493510186672, "lr": 0.0003, "grad_norm": 0.16193281060712603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21147335635, "tokens_per_second": 150723.33504598052, "data_wait_s": 1.1255997378611937e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 144600, "loss": 0.6043801829218864, "lr": 0.0003, "grad_norm": 0.13317204782334968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21162117074, "tokens_per_second": 150840.41150511632, "data_wait_s": 1.1948999599553645e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd7fc69c4414257b780a840d5e294a3bd639b3e79b64fb877bc29e17e3bd91bc
-size 628890
+oid sha256:b48814c87d7da2335291b9e884a859e97b7fd134c888dbee5bd39809ed2b69f8
+size 631008