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": 76015.70338122072,
-      "required_pages_per_second_upper_estimate": 0.006588848509045377
+      "required_token_positions_per_second": 75714.82959492401,
+      "required_pages_per_second_upper_estimate": 0.006562769531806817
     },
     {
       "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": 9501.96292265259,
-      "required_pages_per_second_upper_estimate": 0.003790475076851999
+      "required_token_positions_per_second": 9464.353699365502,
+      "required_pages_per_second_upper_estimate": 0.0037754721953747812
     },
     {
       "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": 19003.92584530518,
-      "required_pages_per_second_upper_estimate": 0.004396223786639624
+      "required_token_positions_per_second": 18928.707398731003,
+      "required_pages_per_second_upper_estimate": 0.004378823322824134
     },
     {
       "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": 28505.88876795777,
-      "required_pages_per_second_upper_estimate": 0.0017358590671287789
+      "required_token_positions_per_second": 28393.061098096507,
+      "required_pages_per_second_upper_estimate": 0.0017289884539952625
     },
     {
       "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": 28505.88876795777,
-      "required_pages_per_second_upper_estimate": 0.002194462226707922
+      "required_token_positions_per_second": 28393.061098096507,
+      "required_pages_per_second_upper_estimate": 0.002185776440355022
     },
     {
       "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": 28505.88876795777,
-      "required_pages_per_second_upper_estimate": 0.35394773542542896
+      "required_token_positions_per_second": 28393.061098096507,
+      "required_pages_per_second_upper_estimate": 0.35254679337567213
     }
   ],
   "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": 0.13049916734360067,
-    "producer_p95_seconds": 0.4323297500013723,
-    "gpu_update_seconds": 0.9699048580824434,
-    "producer_batches_per_second": 7.662884142141902,
-    "gpu_batches_per_second": 1.0310289629614355,
-    "producer_headroom": 7.432268556386348,
+    "producer_batch_seconds": 0.1014063181797269,
+    "producer_p95_seconds": 0.4283214590032003,
+    "gpu_update_seconds": 0.9737590429040969,
+    "producer_batches_per_second": 9.861318485379341,
+    "gpu_batches_per_second": 1.0269481010596249,
+    "producer_headroom": 9.602548050095466,
     "sustainable": true,
-    "buffer_seconds": 60.13410120111149,
+    "buffer_seconds": 60.37306066005401,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 558224.3520874179,
-    "upload_bytes_per_second": 5826233.738404539,
+    "generation_bytes_per_second": 562525.9849288329,
+    "upload_bytes_per_second": 5770932.083898797,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.78195116011526,
+    "estimated_full_shards_per_hour": 60.35249071549768,
     "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": 1.1965257189949625,
-      "bytes": 6621394,
-      "pages": 31,
-      "shards": 4
-    },
     {
       "seconds": 1.1455287000062526,
       "bytes": 6484150,
@@ -562,6 +556,12 @@
       "bytes": 5975632,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9424437819980085,
+      "bytes": 1242871,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -414,3 +414,4 @@
 {"step": 156400, "loss": 0.5990195218473673, "lr": 0.0003, "grad_norm": 0.14261835365343223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22889448533, "tokens_per_second": 149441.1965664636, "data_wait_s": 1.4862998796161264e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156500, "loss": 0.6321857068687677, "lr": 0.0003, "grad_norm": 0.1478285585843141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22904209759, "tokens_per_second": 150204.87217094467, "data_wait_s": 1.2269003491383046e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156600, "loss": 0.6383127085864544, "lr": 0.0003, "grad_norm": 0.1486443915818143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22918824594, "tokens_per_second": 148071.14429043987, "data_wait_s": 7.271002687048167e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 156700, "loss": 0.6271420363336802, "lr": 0.0003, "grad_norm": 0.14207540156982212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22933594966, "tokens_per_second": 149792.27462428136, "data_wait_s": 1.5994002751540393e-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:ff20442b0d23dcfaaa3a5235892480f1e04205bd8e639f369da571f4425598e1
-size 887517
+oid sha256:8b82091f53e40360540e2cfbbcd38ac1fbbab67b12a9fbea7953138b3087b963
+size 889635