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": 74928.4643370199,
-      "required_pages_per_second_upper_estimate": 0.006494609384276214
+      "required_token_positions_per_second": 75618.73176605994,
+      "required_pages_per_second_upper_estimate": 0.0065544400155585995
     },
     {
       "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": 9366.058042127488,
-      "required_pages_per_second_upper_estimate": 0.003736260588051495
+      "required_token_positions_per_second": 9452.341470757492,
+      "required_pages_per_second_upper_estimate": 0.0037706803377842235
     },
     {
       "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": 18732.116084254976,
-      "required_pages_per_second_upper_estimate": 0.004333345382109071
+      "required_token_positions_per_second": 18904.682941514984,
+      "required_pages_per_second_upper_estimate": 0.004373265687463214
     },
     {
       "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": 28098.174126382466,
-      "required_pages_per_second_upper_estimate": 0.0017110313845702478
+      "required_token_positions_per_second": 28357.024412272476,
+      "required_pages_per_second_upper_estimate": 0.0017267940088984575
     },
     {
       "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": 28098.174126382466,
-      "required_pages_per_second_upper_estimate": 0.002163075224973092
+      "required_token_positions_per_second": 28357.024412272476,
+      "required_pages_per_second_upper_estimate": 0.0021830022365243573
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28098.174126382466,
-      "required_pages_per_second_upper_estimate": 0.34888528411019115
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28357.024412272476,
+      "required_pages_per_second_upper_estimate": 0.3520993383447667
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.2620651484533312,
-    "producer_p95_seconds": 0.3445284340014041,
-    "gpu_update_seconds": 0.9839785274175599,
-    "producer_batches_per_second": 3.8158450518958684,
-    "gpu_batches_per_second": 1.0162823396405696,
-    "producer_headroom": 3.754709595018079,
+    "producer_batch_seconds": 0.09483412594519791,
+    "producer_p95_seconds": 0.3117730419980944,
+    "gpu_update_seconds": 0.9749965157851462,
+    "producer_batches_per_second": 10.544727333469316,
+    "gpu_batches_per_second": 1.0256446908374015,
+    "producer_headroom": 10.281072410036979,
     "sustainable": true,
-    "buffer_seconds": 60.02269017247115,
+    "buffer_seconds": 60.449783978679065,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 502419.6096617398,
-    "upload_bytes_per_second": 4954148.618755257,
+    "generation_bytes_per_second": 504531.5358555767,
+    "upload_bytes_per_second": 4996815.354316375,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.8075525034823,
+    "estimated_full_shards_per_hour": 108.26072270155406,
     "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": 1.004709222001111,
-      "bytes": 6095634,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0117634249982075,
       "bytes": 2615219,
@@ -562,6 +556,12 @@
       "bytes": 7222759,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.2303334819989686,
+      "bytes": 10801054,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -289,3 +289,4 @@
 {"step": 383900, "loss": 0.6027638502418995, "lr": 0.0003, "grad_norm": 0.14847226563170499, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56192284609, "tokens_per_second": 149026.94777243093, "data_wait_s": 1.3251999916974455e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 384000, "loss": 0.6020997744053602, "lr": 0.0003, "grad_norm": 0.16682139343107447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56206755170, "tokens_per_second": 147282.2652653989, "data_wait_s": 1.4498997188638896e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 384100, "loss": 0.6103178188204765, "lr": 0.0003, "grad_norm": 0.1743308913183836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56221211483, "tokens_per_second": 147885.76171274134, "data_wait_s": 9.47399894357659e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 384200, "loss": 0.5748583097010851, "lr": 0.0003, "grad_norm": 0.14552985249548286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56235902421, "tokens_per_second": 150183.19092437477, "data_wait_s": 1.3942000805400312e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:061d70a7e6b040d41d1a0f408f930704d1c8c02ac85c263ec8600a45fd068876
-size 620418
+oid sha256:f68b2e44562d160c3feec4eab92b233322ab7faef8795e4389a4dd0fc237886d
+size 622536