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": 75817.25976080698,
-      "required_pages_per_second_upper_estimate": 0.005245213950532356
+      "required_token_positions_per_second": 76255.74083373632,
+      "required_pages_per_second_upper_estimate": 0.00527554908857387
     },
     {
       "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": 9477.157470100872,
-      "required_pages_per_second_upper_estimate": 0.003967887927178669
+      "required_token_positions_per_second": 9531.96760421704,
+      "required_pages_per_second_upper_estimate": 0.003990835785767355
     },
     {
       "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": 18954.314940201744,
-      "required_pages_per_second_upper_estimate": 0.003947683943144995
+      "required_token_positions_per_second": 19063.93520843408,
+      "required_pages_per_second_upper_estimate": 0.003970514954136914
     },
     {
       "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": 28431.472410302613,
-      "required_pages_per_second_upper_estimate": 0.0017044233586529275
+      "required_token_positions_per_second": 28595.90281265112,
+      "required_pages_per_second_upper_estimate": 0.0017142807102030334
     },
     {
       "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": 28431.472410302613,
-      "required_pages_per_second_upper_estimate": 0.0022104219230703087
+      "required_token_positions_per_second": 28595.90281265112,
+      "required_pages_per_second_upper_estimate": 0.0022232056636000007
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28431.472410302613,
-      "required_pages_per_second_upper_estimate": 0.3646884007427125
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28595.90281265112,
+      "required_pages_per_second_upper_estimate": 0.3667975373924285
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.2505145567578211,
-    "producer_p95_seconds": 0.33461153600364923,
-    "gpu_update_seconds": 0.9724434809778367,
-    "producer_batches_per_second": 3.991784002263493,
-    "gpu_batches_per_second": 1.0283373990994868,
-    "producer_headroom": 3.881784330472752,
+    "producer_batch_seconds": 0.5371469137189138,
+    "producer_p95_seconds": 0.3494133589992998,
+    "gpu_update_seconds": 0.9668517962569184,
+    "producer_batches_per_second": 1.8616880679375827,
+    "gpu_batches_per_second": 1.0342846792770226,
+    "producer_headroom": 1.7999764525555237,
     "sustainable": true,
-    "buffer_seconds": 60.291495820625876,
+    "buffer_seconds": 60.91166316418586,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 476663.78976127703,
-    "upload_bytes_per_second": 5774507.711412045,
+    "generation_bytes_per_second": 450021.99573681434,
+    "upload_bytes_per_second": 5802477.509922484,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.28095311764463,
+    "estimated_full_shards_per_hour": 96.56424430921862,
     "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": 0.9538209790007386,
-      "bytes": 2685432,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 0.9428314830001909,
       "bytes": 5761904,
@@ -562,6 +556,12 @@
       "bytes": 6031085,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.012582071001816,
+      "bytes": 5087118,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -327,3 +327,4 @@
 {"step": 517700, "loss": 0.567682683467865, "lr": 0.0003, "grad_norm": 0.15705088020636113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75773299753, "tokens_per_second": 152709.9012266064, "data_wait_s": 1.7655998817645013e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 517800, "loss": 0.5665090084075928, "lr": 0.0003, "grad_norm": 0.13725582150416302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75788025463, "tokens_per_second": 152683.27364554495, "data_wait_s": 1.7523001588415354e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 517900, "loss": 0.6210393905639648, "lr": 0.0003, "grad_norm": 0.17097218424655108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75802622273, "tokens_per_second": 147137.5662933031, "data_wait_s": 1.4174998796079308e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 518000, "loss": 0.5508511066436768, "lr": 0.0003, "grad_norm": 0.1640645318245509, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75817400631, "tokens_per_second": 153574.79664636852, "data_wait_s": 1.699499989626929e-05, "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 sha256:6c480a2831ae8e019b801f7c290e992530c81a0cff714d34ee4b98013f3a5e93
-size 701685
+oid sha256:4d10f7cc46b6e3dad4f5886a23dc2bc7229d2fdae6429496e75ef6571881e37e
+size 703803